Remote sensing of vegetation fires and its contribution to ... · Remote sensing is the science and...

REMOTE SENSING OF VEGETATION FIRES AND ITS CONTRIBUTION TO A FlRE MANAGEMENT INFORMATION SYSTEM

Stephane I? Flasse Simon N. Trigg Pietro N. Ceccato Anita H. Perryman Andrew T Hudak Mark W. Thompson Bruce H. Brockett Moussa Drame Tim Ntabeni Philip E. Frost Tobias Landmann Johan L. le Roux

8.1 BACKGROUND

In the last decade, research has proven that re-

mote sensing can provide very useful support to

fire managers. This chapter provides an overview

of the types of information remote sensing can

provide to the fire community. First, i t considers

fire management information needs in the con-

text of a fire management information system.

An introduction to remote sensing then precedes

a description of fire information obtainable from

remote sensing data (such as vegetation status,

active fire detection and burned areas assess-

ment). Finally, operational examples in five

African countries illustrate the practical use of

remotely sensed fire information.

8.2 FlRE MANAGEMENT AND

INFORMATION NEEDS

As indicated in previous chapters, fire manage-

ment usually comprises activities designed to

control the frequency, area, intensity or impact

of fire. These activities are undertaken in differ-

ent institutional, economic, social, environmental

and geographical contexts, as well as at different

scales, from local to national. The range of fire

management activities also varies considerably

according to the management issues at stake, as

well as the available means and capacity to act.

Whatever the level, effective fire management

requires reliable information upon which to base

appropriate decisions and actions. Information will

Remote Sensing of Vegetation Fires

policies (long-term)

$... . .. . . ,

Fire Management 0 bjectives

A

information

27.0s

, . . . . . .. ..') Monitoring & Evaluation

% Strategies &

Operational Fire

Res. Allocations Management

(short-lerm) &

Research d

be required at many different stages of this fire

management system. To illustrate this, we con-

sider a typical and generic description of a fire

"management loop", as provided in Figure 8.1.

Fire management objectives result from fire

related "knowledge". For example, they may

relate to sound ecological reasons for pre-

scribed burning in a particular land manage-

ment context, or to frequent, uncontrolled

fires threatening valuable natural or human

resources. Whatever the issues, appropriate

objectives require scientific knowledge (such

as fire impact on ecosystems components,

such as soil and vegetation), as well as up-to-

date monitoring information (such as vegeta-

tion status, fire locations, land use, socio-

economic context, etc.).

Policies, generally at a national and govern-

mental level, provide the official or legal long-

term framework (e.g. five to ten years) to

undertake actions. A proper documentation

of different fire issues, and their evolution,

will allow their integration into appropriate

policies, whether specific to fire management,

or complementary to other policies in areas

such as forestry, rangeland, biodiversity, land

tenure, etc.

Strategies are found at all levels of fire manage-

ment. They provide a shorter-term framework

(e.g. one to five years) to prioritise fire manage-

ment activities. They involve the development

of a clear set of objectives and a clear set of

activities to achieve these objectives. They

may also include research and training inputs

required, in order to build capacity and to

answer specific questions needed to improve

fire management. The chosen strategy will

result from a trade-off between priority fire

management objectives and the available

capacity to act (e.g. institutional framework,

budget, staff, etc.), and will lead towards a

better allocation of resources for fire manage-

ment operations to achieve specific objectives.

One example in achieving an objective of con-

serving biotic diversity may be the implemen-

tation of a patch-mosaic burning system

(Brockett et al., 200 1 ) instead of a prescribed

block burning system, based on an assump-

tion that the former should better promote

biodiversity in the long-term than the latter

(Parr & Brockett, 1999). This strategy requires

the implementation of early season fires to

reduce the size of later season fires. The

knowledge of population movements, new

settlements or a coming El Niiio season,

should help focus the resources usage, as

these factors might influence the proportion

as well as the locations of area burned.

Another strategy may be to prioritise the grad-

ing of fire lines earlier than usual based on

information on high biomass accumulation.

Figure 8.1. Typical fire "management loop"

Wildland Fire Management Handbook for Sub-Sahara Africa

However, whatever the strategies, they need

to be based on reliable information.

Operational fire management concerns the

implementation of the strategy. Daily activities

will also be most effective if based on reliable

and up-to-date information. For example,

an accurate knowledge of fire frequency,

fuel load, fuel status and meteorological

conditions across the management area will

help to inform the choice and timing of areas

for ignition within a prescribed burning

programme; early detection of active fires in

relation to their potential impact will help

prioritise the activities of fire fighting teams.

Research activities may require a range of

studies - from long-term to short-termlone-

off - in order to answer specific questions of

concern to improving fire management.

Monitoring and evaluation activities are essen-

tial t o close the "management loop". They

allow the assessment of the effectiveness of

different strategies, to document the current

situation, and to learn from the past in order

to adapt and improve knowledge and manage-

ment activities for the next loop.

Repeating the loop is also an essential part of

management, in order to evolve with the

natural, economic, and societal changes. Up-

dated information will always be required to

act appropriately.

A Fire Management lnformation System (FMIS)

is an important tool to support integrated fire

management. It allows for incorporating infor-

mation and knowledge from various sources and

integrating them into thematic information in

direct support of specific decisions. FMIS can in-

clude information such as:

Fire events over the years (e.g. where, when

and how often have areas burned).

lnformation that may be related to the fire

events (e.g. what vegetation was burned,

ecological knowledge obtained in the field,

desired fire regimes, areas where fires are

acceptable/unacceptable (under management

or not), why fires are set, attitudes of differ-

ent people towards fire and fire prevention,

population density, meteorological data,

vegetation status, economical assets).

Ancillary information (e.g. roads and river net-

works, administrative boundaries, protected

areas, concessions, villages, fire towers, fire

fighting units).

Modelling tools, e.g. fire prescription models,

fire danger models and fire spread models.

Fire is seen as an efficient tool in the management

of (often) large areas of land (Bond & Van Wilgen,

1996). However, whilst field observations will

always be a vital part of fire management, the very

size of the areas in question often means that field

observation alone cannot provide sufficient infor-

mation with sufficient accuracy and regularity to

provide a reliable basis for fire management. Such

problems are compounded in countries and regions

where resources and local staff are particularly

constrained. Many studies have demonstrated the

potential usefulness of remote sensing techniques

for monitoring the Earth's surface and providing

fire related information in particular (e.g. Kaufman

et al., 1990; Pereira et al., 2000).


Due to a high correlation between variations

cissewed from remote sensors and variations on

the Earth's surface (Congalton & Green, 1998),

remotely sensed data provide an excellent basis

for monitoring parameters of interest to fire

managers, such as biomass, vegetation status, the

occurrence of active fires and the delineation of

areas that burn. It works because the Earth's

surface reflects light and emits energy differently

according to its land cover type, status, quantity and

several other factors. The technology can give the

geographical location of any point of an image,

therefore allowing its combination with other

geographic information such as roads, fire units,

protected forest, plantations, villages and other

fire-related information, as well as the cross-

comparison of images taken at different times

within and across seasons.

The benefits that remotely sensed data provide

to fire management include:

it is often less expensive and faster than ob-

taining the same information on the ground

over large areas.

It permits the capturing of data across a wider

range of the electromagnetic spectrum than

can be seen by humans. This can allow the

extraction of a wider range of fire-related

information.

Observations are spatially comprehensive.

They cover large areas of territory (e.g. the

whole of Ethiopia at once), including areas

that are remote and difficult to access by land.

In the case of satellite observations, observa-

tions are regular (e.g. daily), allowing for

frequent updates of the situation.

Because the satellite orbits Earth continuously,

observations are reliable, systematic and

objective (i.e. the same place can be imaged

repeatedly with the same sensor).

8.3 REMOTE SENSING DATA:

INTRODUCTION

8.3.1 A Short Introduction to Remote Sensing

One of the simplest, broad definitions of remote

sensing is that given by Lillesand and Kiefer (2000):

Remote sensing is the science and art of

obtaining information about an object, area,

or phenomenon through the analysis of data

acquired by a device that is not in contact

with the object, area,

investigation.

or phenomenon under

You are therefore using

read these words! Your

remote sensing as you

eyes are sensing varia-

tions in light from the page and your brain is

interpreting this "data" so that you can under-

stand the information that the words convey

(Lillesand & Kiefer, 2000). Other definitions add

that the information is usually derived about the

Earth's land, water and atmosphere from images

acquired at a distance, based on the measurement

of electromagnetic energy from these features

(Campbell, 1987).

In the context of Earth observation remote

sensing, an image is generally a picture received

from a satellite or an airborne sensor. Digital

images from satellite remote sensing are useful

for fire monitoring because they:


Allow low cost, rapid and regular coverage of

the often extensive and inaccessible areas

affected by fire.

Permit capture of types of data that humans

cannot sense, such as the near-infrared and

thermal part of the electromagnetic spec-

trum, which may provide additional useful

information.

Here we briefly introduce the general character-

istics of digital images, mostly from space-borne

sensors, as a potential source of information for

fire management. As different sensors provide

images with different characteristics, we focus

on criteria commonly used to evaluate and com-

pare imagery from different sources. Annexure I

summarises satellite sensors currently provid-

ing data for Africa.

8.3. l . 1 Spatial Resolution

An image may look, at first sight, like a phoeo-

graph. However, enlarging the image reveals tlaat

it is actually made up of many small square blocks,

called pixels (short for picture elements).

All sensors have a limit on how small an object

on the earth's surface can be and can still be seen

by a sensor. This limit is known as the spatial

resolution and is related to the image pixel size.

The 30 m spatial resolution of the Landsat-Wl

image, used in Figure 8.2, renders a detailed view

of a burned area, with the complex perimeter and

unburned islands of vegetation clearly visible.

The low spatial resolution NOAA-AVHRR

sensor uses a pixel size of I. I km, which means

that most objects smaller than I km cannot be

detected reliably (with active fires being an

important exception). Figure 8.3 shows how

the same burned area was mapped from $PI

Figure 8.2. In the overview image (A), a burned area is clearly evident in shades of medium to dark blue. Unburned

vegetation appears green. With increasing magnification (B), the image appears more "grainy", until in (C), individual pixels - that make up the image - can be seen. The image is made from TM data with a spatial resolution of 30 m. The intensig, or

brightness, with which each pixel is displayed, is proportional to the average brightness, or radiance, measured electronically

over the ground area corresponding to each pixel.

I 62


m d AVHRR data. The images reveal the degree

nF simplification inherent at coarse spatial reso-

lution.

8.3.2 Swath Width

5ensors on polar orbiting platforms cover a

"'swath" or "strip" of the Earth's surface, with

i21e width of the swath, and hence the width of

::he image, depending on the particular sensor. In

general, broad-swath imagery (e.g. 2700 km

wide) is well adapted to the frequent observation

cd large areas, but at the expense of spatial detail,

while narrow-swath imagery (e.g. 185 km wide)

provides the spatial detail but is available less

frequently.

9.3.3 Temporal Resolution

The frequency with which a satellite is able to take

an image of a particular area of ground is also im-

pwtant. The time interval between images is called

t1.e return period. The shortest reliable return

period is known as the temporal resolution of the

sensor. This usually varies between 15 minutes to

over 30 days, depending on the satellite.

The temporal resolution is largely determined

by the orbit characteristics of the satellite, but

the spatial resolution of the sensor will also affect

this. For example, the NOAA AVHRR sensor

scans a continuous swath 2700 km wide and can

image the entire earth surface twice per day, but

at a spatial resolution of only I. I km. A SPOT

sensor covers a swath around 60 km wide with a

spatial resolution down to 3 m, but the narrow

swath means that it takes 26 days to image all of

the Earth and therefore one place is only re-

visited every 26 days (see Annexure I).

8.3.4 Spectral Resolution

The human eye can see many different colours

that, taken together, make up visible light. Visible

light is only one of many forms of electromag-

netic energy. Radio waves, X-rays and ultraviolet

rays are other familiar forms. All electromagnetic

energy travels in waves at the speed of light. The

distance from one wave peak to the next is called

the wavelength. The electromagnetic spectrum

is divided up according to wavelength (usually

measured in micrometers - mm), although there

Figure 8.3. The same burned area, (A) mapped from T M data with a spatial resolution of 30 m and (B) mapped from AVI-iRR data with a spatial resolution of approximately I . I km (at best). Although the burned area is approximately the same shape in both pictures, the AVHRR representation is highly simplified compared to using TM, illustrating the loss of detail at lower spatial resolutions.

W~ldland Fire Management Handbook for Sub-Sahara Africa

are no clear-cut dividing lines between the

different regions. Satellite sensors are sensitive

to a much wider range of wavelengths than that

of visible light. Sensors effectively "see" at wave-

lengths that are invisible to the eye, and this

often allows more information to be obtained

about objects than would be possible by simply

loolcing at them.

Objects reflect and emit different amounts of

radiation at different wavelengths. In the visible

to mid-infrared, this response is measured using

reflectance. In practice, satellite sensors usually

provide each image in a number of different bands

or channels. Each band is sensitive to electro-

magnetic radiation over a restricted range of

wavelengths. By strict definition, the narrowness

of this range gives the spectral resolution of the

band. However, in the context of satellite remote

sensing, spectral resolution can be more usefully

interpreted as the particular band used. The sen-

sor makes measurements of the total response

across the particular band used. N o more pre-

cise reading can be made by this sensor within

the band.

Comparing reflectance spectra of different

surfaces can help to determine which bands are

most appropriate for looking at each cover type.

Figure 8.2 shows an example of reflectance spec-

tra for a burned surface, green shrub and senes-

cent grass. The approximate wavelength inter-

vals (blue, green, red, near infrared [NIR], short

mid-infrared [SMIR] and long mid-infrared

[LMIR]) are also shown. It is possible to distin-

guish both vegetation types from the burned sur-

face in the near infrared, because the reflectance

of the burned surface is low and the reflectance

of the vegetation is high. Hence they will appear

dark and light respectively on a near infrared image

band. At visible wavelengths, the two vegetsiion

spectra (particularly shrub) are similar to the

burned surface, suggesting that visible bands do

not provide good contrast between burned and

unburned vegetation. In the SMIR, only grass con-

trasts strongly with the burned surface, whilst in

the LMIR, only shrub has good contrast.

Clearly, discrimination between surfaces de-

pends on the band used. In fact, sensors that take

measurements in few broad bands offer less

potential information than sensors that measure

EM energy in many bands positioned over a wider

range of wavelengths. For example, panchromatic

air photos (i.e. sensitive to all colours) are sensi-

tive to light reflected from the surface (approxi-

mately analogous to having one band in the

visible). Using these photos, some burned areas

can only be interpreted reliably up to three days

after the fire. In contrast, data from the Landsat-

TM sensor, provided in seven bands over a much

wider spectral range, can identify the same

burned area months after burning. Similarly, other

combinations of spectral bands can be used to

Figure 8.4. Spectral response (variation of reflectance with wavelength) of a burned surface, compared to senescent grass and green shrub. The approximate wavelength intervals are also marked with dashed lines.


derive other fire-related information such

active fires and fire risk.

8,3.5 Cost

Data costs vary from free, unlimited access to

as

all

available images (as is the case with AVHRR data,

so long as the necessary receiving equipment is

in place, and for MODIS), to costs of well over

one thousand US dollars for each image acquired.

In general, prices increase with spatial resolution.

Low to moderate spatial resolution, free data

(e.g., NOAA-AVHRR and MODIS) can be very

useful for fire management.

8.3.6 Operational vs. Research

Satellite Programmes

Operational satellite programmes are organised

to guarantee the routine availability of particular

kinds of remotely sensed data from the same

t'jpe of instrument over extended or indefinite

time periods. As such, they offer a very impor-

tant resource for comparing patterns and trends

in surface cover and processes between years.

For example, the NOAA-AVHRR has provided

data operationally since 1979, which has been used

in studies of global change, and is a valuable re-

source for studying fire patterns over the years.

Research satellite programmes do not place

the same guarantees on prolonged availability of

data, and are primarily aimed at demonstrating

or using improved technology to provide better

information. As such, they are also important

potential sources of improved fire management

information, but there are less guarantees as to

how long into the future the data will remain

available.

8.3.7 Data Access

Remotely sensed data has in general become

easier, cheaper and quicker to access through

time. Initially, all data had to be ordered from

large, centralised receiving stations, usually far

from the institutions requiring the data. Raw data

was usually delivered on tape, or as hardcopy,

which could mean having to wait several weeks

to obtain it. The advent and rapid development of

personal computers, combined with improve-

ments in receiving hardware, resulted in PC-

based receivers that allow local institutions to

access low spatial resolution imagery themselves,

in near-real time. For example, LARST (Local

Application of Remote Sensing Techniques) re-

ceiving units provided direct access to AVHRR or

Meteosat data in many organisations in over 40

different countries (Williams, 1999; Downey,

1994). Further advances in technology resulted

in portable, high specification receiving stations

capable of allowing local institutions to collect

their own images directly from high spatial reso-

lution sensors such as Landsat-TM, ERS-SAR and

SPOT-HRVIR (Downey, 2000).

With the advent of the internet, organisations

who launch satellites are increasingly providing

images and other products online, for rapid

access by end-users. For example, fire and other

data from the MODlS sensor is obtainable over

the internet free of charge and data from the

operational SPOT VEGETATION sensor is also

available online.

At the time of writing, most high spatial reso-

lution satellite data is still received through a

network of few grounds stations, and their distri-

bution organised centrally.


Clearly, choosing a sensor and route to pro-

vide particular fire management information will

require careful consideration of the above aspects,

to identify a data source suitable for providing the

desired information of the area of interest with

sufficient detail, accuracy, regularity and economy,

t o support specific fire management objectives.

Some of these issues are explored further in the

section on burned area products (8.4.3).

8.3.8 Other Considerations

It is worth mentioning some additional characteris-

tics of remotely sensed data that the fire manager

will need to bear in mind. Thick cloud cover will

obscure the surface in most bands used in op-

erational remote sensing for fire (only radar

observation can go through clouds). The same is

valid for thick smoke (except at the mid-infrared).

Centralised receiving stations usually provide

browse products of the images on offer that can be

visually inspected for cloud and smoke, so that

cloud-free images can be identified and ordered.

The accuracy of maps made from remotely

sensed data is variable and depends on many fac-

tors, and quality control is therefore important

at all stages of map production. It is extremely

important to choose a data source that will register

the different features t o be mapped w i t h

distinctly different levels of electromagnetic re-

sponse. Spatial, temporal and spectral resolutions

are all important in this regard. Secondly, having

identified an appropriate data source, a robust

method must be chosen and applied to extract

the desired information and deliver the final map.

Uncertainty in the accuracy of maps derived from

remotely sensed data generally increases with

decreased spatial resolution, spectral resolution

and longer return periods. As we have seen, the

accuracy of maps made from low spatial resoiu-

tion data is inherently limited by the low spatial

precision of the raw data.

Realisation of the full potential of any maps

made from remotely sensed data therefore re-

quires the accuracy of the map to be assessed.

This can be done quite simply by collecting a sam-

ple of reference data (assumed t o be true) at

representative locations, which are then com-

pared with the same locations on the map. The

overall accuracy can then be estimated, as well as

other measures of accuracy, that are of direct

interest to the producer and users of the map.

This can then help t o ensure the adequacy of the

maps (and hence the data source and methods

used) for providing the required management

information. Congalton and Green ( 1999) provide

a comprehensive introduction to both the princi-

ples and practices of assessing the accuracy of

remotely sensed data.

8.4 REMOTE SENSING PRODUCTS

FOR FIRE MANAGEMENT

8.4.1 Introduction

Remote sensing data can assist fire management

at three stages relative to fire occurrence:

Before the fire: fuel load, vegetation status (e.g.

degree of curing, moisture content) and rain-

fall.

During the fire: near real-time location of

active fires.

After the fire: assessment of burned areas.


Figure 8.5 gives a basic idea of how fire activity at

i$-:e surface of the Earth is seen from space. In

d;is case, using a thermal image that is presented

so that hot areas appear relatively bright and cooler

ai-cas are relatively dark. As one might expect,

active fire fronts and burned areas stand out as

bright features that contrast well with cooler

aEas such as smoke and unburned vegetation.

from this simple example, we might conclude

t R b the extraction of active fires, burned areas

and other fire-related information from remotely

sei>sed data should be straightforward. However,

in i-adity, it is often far from trivial. In our example,

are ;:he observed bright areas in Figure 8.5 defi-

nitely active fires or are they burned areas, and how

dc w e distinguish between the two? Are cold areas

smoke or vegetation or even water? Are the diffe-

retit .features best distinguished using a thermal

imzge alone? -- I he sensor on board a satellite platform (or a

camzra on board an aircraft) only observes electro-

rnzgimic (EM) radiation coming from the surface

of ths Earth. Proper extraction of adequate

information requires methods that are based on

the knowledge of how fire-related features

impose variations in radiation quantities that are

measurable by remote sensors. The observed

surface radiation can come from reflected sun-

light or from emission by the surface itself." For

example, a fire will be hot and reflective, whereas

water will be relatively cold and unreflective, both

leading to different quantities of radiation being

measured by the sensor. Using these differences

and variations, digital processing methods, known

as algorithms, can be designed to extract (from

the signal) information in terms of active fires,

burned areas, fuel load, vegetation moisture and

rainfall. If appropriate methods for digital image

processing are unavailable, images can be inter-

preted visually using similar techniques to air

photo interpretation, with the interpreted areas

either digitised from a computer-displayed im-

age, or drawn on hardcopy.

It is important to realise that the accuracy of

fire information obtained from remote sensing

will vary considerably, depending both on the char-

acteristics of the sensor used to obtain the raw

data, and on the precision or appropriateness of

the algorithm or visual interpretation used to

transform the raw data into fire information. It is

therefore important that measures are taken to

assess the quality or accuracy of any information

obtained from remote sensing. This is a vital step

to ensure that the best information extraction

techniques are chosen, to allow accuracy to be

improved where necessary, or to at least ensure

that any inherent limitations are accounted for

realistically when making decisions based on the

remotely sensed information. In short, the right

Figma 8.5. Thermal image from the AVHRR sensor, over northern Botswana. White is hot, and black is cold.

" In LIX case of "active" remote sensing (such as some radar systems), sensors actually measure the quantity of radiation,

initlzliy sent by the sensor itself, which bounces back from the earth surface. These are so far not used very much in the field of fire monitor~ng, and are not detailed here.


decisions can only be assured if the accuracy of

the remote sensing technology is quantified and

where necessary accounted for.

The following sections of this chapter describe

various remote sensing products useful t o fire

management, covering their use and the method

for their extraction.

8.4.2 Active Fires

Active fires can be detected from satellite data

because fire fronts are very hot and emit large

amounts of energy that can be observed by thermal

sensors onboard satellites or aeroplanes. The

identification of fires in an image is now relatively

well mastered, and remaining limitations are

mostly due to the sensor in itself. The basic active

fire product is a l is t of locations (latitude and

longitude) corresponding to pixels detected as

having an intense source of heat in the area of

land they cover.

8.4.2.1 Active Fire Product

in Fire Management

Once integrated into a fire information system,

the list of fire locations can be used in two main

ways:

In near-real time, to prioritise resources for

fire fighting. Within minutes of the satellite

overpass, the fire manager can locate active

fires on the territory of responsibility. Intro-

duced into the fire information system, the

importance of a fire can be considered. For

example, a fire in an agricultural area, at the

time of land preparation, may mean a con-

trolled good fire, presenting no risk. On the

other hand, an unexpected fire near a coffee

or a young palm tree plantation, for example,

may be more important to tackle. Fire loca-

tions can also be used, on a daily basis, to

monitor, for example, that planned prescribed

burning is actually taking place.

As post-fire information, the active fire product

can be used in several ways. Firstly, it can

support a policing role. When officers go out

in the field to see farmers and villagers, fire

maps can provide strong evidence that there

is official monitoring and therefore can be

useful to promote alternative or preferred

fire practices. Secondly, active fire products

can be used to document fire activity in a park,

over a municipality or over a whole country.

They have been used in this way since the

mid-1980s. Due to the nature of active fire

observation (see further discussion) as well

as scientific progress, the direct mapping of

burned areas is increasingly seen as a way of

providing more complete fire figures. Never-

theless, active fire locations still remain

valuable and complementary products in, for

example:

Documenting the extent of individual fire

fronts and the size of fires that contribute to

the burned area mosaic.

Documenting trends over the years.

Documenting the type of fires according to

the vegetation in which they occur.

Identifying areas of particular human pressure

on natural forest.

Monitoring and evaluating fire strategies (pre-

scribed burning, awareness campaigns, etc.).


8.4.2.2 Operational active fire products

i here are a number of satellite and airborne

remote sensing systems which can contribute to

.-ire monitoring from space, including NOAA-

AVI-IRR, Landsat-TM and MSS, SPOT, GOES,

i3MSe ERS-ATSR, JERS and MODIS. The tempo-

[-a!, spectral and spatial characteristics of these

instruments provide a wide range of sensing

capabilities (Justice et al., 1993) and some of them

i;ave been shown to be well adapted to fire

detection applications. However, the usefulness

c; operational near real-time fire detection from

space i s obviously very much dependent on

observation frequency.

High spatial resolution satellites, such as

"Lndsat and SPOT, can contribute t o fire moni-

toring, but their cost, their centralised receiving

~Lations and especially their low temporal reso-

lution, limit their use on an operational basis.

P'leceorological satellites are more appropriate

because of their high repetition coverage. The

P-ieteosat geostationary satellite series* covers

Ah-ica and Europe, and provides images every 30

minutes (Meteosat Second Generation satellite,

launched in mid-2002, provides an image every

15 minutes, with improved channels for fire infor-

m~tion). The polar orbiting NOAA series acquires

ilixges over the same area every 1 2 hours by the

same satellite, and covers the entire world. There

are early afternoon and early morning passes

mailable, as there are two operational satellites.

High temporal frequency is especially useful if the

data can be acquired, analysed and disseminated

in mar real-time. Satellites such as NOAA and

Meteosat broadcast their data continuously and

01-11:' require small receiving stations. A number

of these stations are distributed all over the world.

Local acquisition of data free of charge, analysis

in situ, and fast dissemination of fire information

is possible with these two satellite series (e.g.

Jacques de Dixmude et al., 1999).

Several authors have developed algorithms for

active fire detection with AVHRR data. The reader

will find agood review and further details on these

algorithms in Martin et al. (1999). They are all

based on using AVHRR mid-infrared channel, most

suited to be sensitive to fire front temperature

level.

There are many factors that can affect the de-

tection, such as cloud and smoke, hot soil and sun

glint on water. Flasse and Ceccato (1 996) devel-

oped a contextual method designed to be robust

and automatic, for operational use. It is used op-

erationally in several tropical countries (e.g. Flasse

et al., 1998). It has also been the basis for global

fire detection activities such as the IGBP-Global

Fire Product and the World Fire Web of the joint

Research Centre (see http://www. gvm. jrc.itl

TEM/wfw/wfw.htm).

Up to now, it is essentially NOAA-AVHRRthat

has provided long-term, continuous operational

satellite-based systems, allowing low-cost direct

Figure 8.6. Fire pixel interpretation.

$' i t s sister, covering the Americas, is the GOES series.


reception and near-real-time fire information

over Africa. However, when the documentation

of the fire activity does not require long-term

and continuous coverage, and when near-real

time is not an issue, other sensors, as mentioned

above, can provide a valuable contribution to prac-

tical studies.

8.4.2.3 Product Interpretation

There are several points that are important to

take into account when interpreting and using

active fire products from AVHRR data. Most of

them are linked to the intrinsic characteristics of

the satellite platform and its sensor. Detection

algorithms are usually set t o minimise the

number of false detections. Consequently, some

fires will also be missed. The main points to under-

stand are described below:

Fire and pixel size. AVHRR was not initially

designed to detect fires. The AVHRR signal

over an active fire saturates quickly, and thus

does not vary very much between small and

large fires. Consequently:

- Very small fires are not detected. Pixel

size conditions the minimum area that

has to be burning to have a signal detect-

able from the satellite. Belward et al.

( 1 993) demonstrated that a bush fire, with

a burning front as small as 50 m, could be

detected by AVHRR I x l km pixel.

- A pixel detected as fire could represent

different situations.

- There could be one or several active fires

in the area covered by the pixel, or the

pixel area could all be covered by a large

fire front, of which the pixel would only

be a part.

Location accuracy. The location of a fire can

only be given within a variable range, which

for AVHRR typically varies between I and 3

km. The term "fire location" refers to the

central latitude and longitude of the fire pixel.

It is easy to understand that - depending on

the fire size and the pixel size as described

above -the central point of the pixel may not

exactly represent the position of the fire, In

addition, errors can also come from the actual

geographical registration accuracy of satellite

image in itself.

Timing. Only those fires that are active at the

time of the satellite overpass will be detected.

Those fires starting after image acquisition

will not be detected until the next image, or

missed if they are extinguished prior to the

acquisition of the next one. While this can be

a constraint for fire fighting, because the

NOAA satellite passes in the afternoon, local

time, corresponding to high fire activity, active

fire products will be representative of the

general fire activity.

Clouds. Although AVHRR channel three can

see active fires through smoke and thin clouds,

fires under thick clouds are not visible from

the satellite.

Finally, it is important to note that products should

be field validated where possible. However, it is

difficult to validate remote sensing products because

of scale issues, as well as the cost associated with

exhaustive validation campaigns. Experience

shows that current algorithms perform well, and


,.I- L. e existing imprecision is usually greatly out-

tf>.,zighed by the advantages of remote sensing

o.r.;ulrvations (large area, repeated coverage, etc.).

I-. :.aver, users should always be aware of these

is::i-;es and, when possible, adjust algorithms for

tli sit- own region.

8 ' Burned Areas

8.4.3.1 Burned Area Product Principles

BI -*-ted areas are detected from remotely sensed

dz z based on three main changes in surface prop-

ei;;es following fire:

6 "kgetation is removed.

Combustion residues are deposited.

During the day, the burned surface is hotter

than surrounding vegetation, with a maximum

contrast in temperature occurring around

i-t7 i d - day.

As ehe above changes remain for some time after

bur-i4ng, a "memory" is held of the affected areas.

This "memory" i s unavailable t o active fire

detection, but enables burned areas to be mapped

d ~ ~ . i i ~ j : entire fire seasons using relatively few re-

rnot~ly sensed images (Eva & Lambin, 1998). The

mairi downside is that, at present, burned area

detedion methods are generally less automated

than active fire-based methods. --.

i he basic burned area product is an image,

which shows burned areas in a different colour

to unburned areas. Burned area products are

usudi? provided in a standard map projection,

So thz-t the geographic coordinates (e.g. latitude1

longi::ude) of any pixel are easily obtained.

8.4.3.2 Burned Area Products

in Fire Management

Integrated into a Fire Management Information

System, burned area products are useful at all

stages of the fire management loop:

Baseline data

Burned area products can provide important base-

line information on fire regimes (i.e. frequency,

season and intensity). Fire frequency maps are

obtained by superimposing burned area maps for

successive years. Seasonal fire maps are produced

using several successive burned area products.

Figure 8.7 shows a time series of burned area

products for Caprivi, north-east Namibia, which

includes parts of Angola to the north, Botswana

Figu:,~ 8.7. Burned area products, showing the progressive accumulation of burned areas during the 1996 fire season in the

C a p w 2nd Kavango regions, north-east Namibia and surrounding areas. The products are based on NOAA AVHRR images

of the 21-ea, which were acquired at regular intervals throughout the f ~ r e season.

ecological information on desired fire regimes, it

is possible to highlight areas where existing

regimes are acceptable, or rather deviating, from

the Intention. This is a powerful tool for developing

fire policies, modelling their outcomes and then

formulating strategies, and for helping to direct

fire management activities.

Refining policy

All the above are then used to refine fire manage-

ment policies. Fire frequency maps can also help

identify areas where high intensity fires are burning

frequently, as foci for field visits to investigate the

causes and the fire effects.

Burned area products

Figure 8.9 is a flow chart of the steps typically

involved in preparing burned area products

(although the flow may not be so linear). This

procedure is important because one must choose

the appropriate technique according to the

product required. The steps are expanded below.

Specify format of the burned area products.

It is first necessary to choose which burned

area products are needed in order to provide

information required by management.

Decide on appropriate scale of mapping.

Scale includes the dimensions of the area that

is to be mapped, the level of detail (or spatial

resolution) that is required, and how often

the map needs to be updated, (i.e. the tem-

poral resolution of the map). In Figure 8.7 the

large area involved ( 1 45 000 km2) meant

that small-scale mapping was the only option.

In Figure 8.8, the area is much smaller at


604 km2, thus large scale-mapping was

attainable. The decision on the level of detail

required for the map (spatial resolution)

should be made carefully in relation to manage-

ment needs. For example, block burning (Du

Plessis, 1997; Stander et al., 1993) results

in relatively large and homogeneous burned

areas (Parr & Brockett, 1999; Brockett et al.,

200 1). High detail is generally not crucial to

map these adequately, and even AVHRR

imagery (with a I. l x I. l km pixel size) can

often yield more accurate results than the

usual field-based method of driving block

perimeters. An interpretation of a comparison

between burned areas mapped using AVHRR

and TM also found that AVHRR was mapping

fires at the scale of the field mapping under-

taken by section rangers in the Kruger

National Park (Hetherington, 1997; 1998).

Data from sensors such as AVHRR (which

can be accessed freely each day using a

relatively low cost, PC-based receiver) and

MODlS (data freely available over the Inter-

net) become attractive choices. In contrast,

fire that is prescribed using a patch mosaic

system results, in numerous small but eco-

logically important burned areas (Parr &

Brockett, 1999). Higher detail is needed to

resolve these accurately, so images from sen-

sors, such as SPOT-HRVIR and Landsat-TM

are required. The downside is the much

higher costs for covering smaller areas, which

means that management will want to know

the minimum number of images needed to

map burned areas each season. The regularity

with which images need to be obtained


I. Choose f ~ r m a i at burned area product, tor exflrnpls - Dire frequency (annual maps post-fire) - Dale-of-burn maps {multi-temporal maps duririg fire season)

- Fire intens~ty - Fire severrty

2. Decide on appropriate scale of mapping: Dimensions of area to be covered

- Level of detail (spstlal resolution) - How often (temporal resolut~on)

3. Seled sttitable source of ~mgery: SPOT HRVIR: Landsat TM, MODIS. SPOTVGT, AVHRR, etc.

4. Decde on appropriate method for mapping burned areas and apply it to the Imagery to produce the des~red pt-oduct.

5. Assess map aceuracy using reference data collected at representative locations.

depends on the product type and the temporal

spacing of images required to ensure that

burned areas are not missed.

Select suitable source of imagery

Having weighed up the requirements of pro-

duct format and scale, the image data source

can be chosen. In making a choice, it is impor-

tant to also confirm that this source will be

adequately sensitive t o the parameter of

interest, perhaps via a pilot study or by con-

ducting a literature review. It should also be

remembered that, for monitoring purposes,

it is important that scale is maintained. Hence

budget constraints are very important con-

siderations in making a final decision.

Decide on appropriate method for mapping

burned areas.

Having chosen an appropriate data source,

the accuracy of burned area mapping will

depend on the method used. Compared to

active fires, burned areas contrast relatively

weakly with unburned vegetation, and so it is

important to choose a robust method that is

sensitive to changes caused by burning, yet

insensitive to changes from other sources of

variation (Trigg & Flasse, 200 1 ). In general,

burned areas that are smaller than the ground

area covered by one image pixel cannot be

detected.

Burned areas are often obvious visually to an

image interpreter because of the superior ability

of the human mind, relative to current computer-

based methods, in recognising spatial patterns.

Edwards et al. ( 1 999) compared five burned area

mapping techniques and found on-screen manual

digitising to be more accurate than automated

image processing techniques. However, the

patchiness of burned areas makes manually digi-

tising them very tedious and subjective. Further

considerations of time, practicality, objectivity and

ability to repeat, make automated analysis tech-

niques preferable for extracting burned areas

from remotely sensed imagery. Most image

processing techniques operate in the spectral

domain, that is, they use differences in the amount

of energy received from burned and unburned

areas in the different spectral bands available

to discern between the two cover types. Visual

interpretation uses both the spectral domain

(manifested as variations in image brightness or

Figure 8.9. Flow diagramme showing the main steps and considerations in the preparation of burned area products.


c o i ~ ~ r ) and the spatial domain (variations in pattern

and texture). Multi-spectral imagery typically in-

cludes bands in the near- to thermal-infrared,

which contain more spectral information indicative

of burned areas than the visible channels (Pereira

& Setzer, 1993; Pereira et al., 1999a; Trigg &

Flasse, 2000). For visual interpretation, any com-

bination of three bands may be displayed, for

example using the red, green and blue colour

guns of a computer screen, although no more

than three bands may be displayed at once. On

the other hand, there is no practical limit to the

number of spectral bands that can be simulta-

neously processed by computer-based methods

to detect burned areas.

Computer-based detection methods are usu-

ally based on identification of one or more of the

physical changes mentioned in the introduction

to this section:

Methods sensitive to vegetation removal

usually use vegetation indices (Vls - simple

algebraic combinations of more than one

band), whose values tend to decrease sharply

after burning, providing a basis for detection.

Historically, NDVl was the most commonly

used VI for detecting burned areas, and it has

been used on all fire-prone continents,

although numerous inherent limitations have

now been described. More recent Vls such as

GEM1 (and its variants) and atmospherically

resistant Vls (ARVls) are increasingly used in

preference to NDVl (Pereira, 1999; Miura et

a!., 1998). Vls are most useful for detecting

burned areas if primarily photosynthesising

vegetation burns (e.g. in pine and evergreen

forests). However, in areas such as grassland,

shrubland and deciduous woodland, wide-

spread vegetation senescence can occur prior

to burning, which can decrease the accuracy

of VI-based detection (Trigg & Flasse, 2000).

Certain land management activities that alter

vegetation abundance (e.g. tree felling) may

also be mistaken for burning using Vls.

Burned surfaces covered by char combustion

residues usually appear much darker than

unburned vegetation, particularly in the near-

infrared (NIR), providing a very good basis for

detection (Trigg & Flasse, 2000). However,

this basis is short-lived in areas where char

is removed rapidly by the wind and rain,

making the burned area brighter and less

distinguishable from unburned vegetation.

Other cover types, such as water, may be

indistinguishable from burned areas in the

NIR, and so bands at other wavelengths are

often needed to help resolve this confusion.

NIR bands are less discriminating in areas

where more efficient combustion results in

bright ash residues that contrast less strongly

with unburned vegetation.

As one might expect, methods that detect

burned areas as hot surfaces use bands in the

thermal infrared (TIR). While generally robust,

thermal-based detection is not possible at

times or in places where surface tempera-

ture exceeds the upper limit that can be

measured by a particular sensor. For example,

AVHRR band three images are useful for

detecting burned areas, but only if surface

temperatures stay below approximately 5 1 "C,

i.e. the highest measurable temperatwe.


In Namibia, un-shaded surface temperatures

usually exceed this limit around mid-day from

August and October, rendering AVHRR band

three images unusable. New sensors, such as

MODIS, can measure much higher tempera-

tures and so avoid this problem of "saturation".

The utility of night-time thermal imagery is

limited due to the poor thermal contrast be-

tween burned and unburned areas found at

night. Another constraint i s that smoke

plumes present cool features that can con-

ceal underlying burned areas at long-thermal

infrared wavelengths.

In practice, burned area detection methods

usually combine spectral bands to provide sensi-

tivity to one or more of the fire-induced changes.

Examples include multi-spectral image classifi-

cation, principle components analysis (Hudak et

al., l998), and spectral indices designed specifically

to detect burned areas (Trigg & Flasse, 2000).

Many of the available methods are reviewed in

Pereira et al. ( 1 999b) and Koutsias et al. ( 1 999).

Detection methods can also be grouped

depending on how many images they use. Single-

image detection is based on the assumption that

all burned areas will be distinguishable in the

spectral domain on just one image. Although one

image is quick and cheap to obtain and process,

several other cover types, such as shaded slopes,

water bodies, urban areas and bare soils may be

indistinguishable from burned areas on imagery

taken on a single date. Some of the confusion may

be resolved by using spectral information from

all of the available spectral bands in the image,

sometimes in conjunction with sophisticated

image transformation techniques (e.g. Koutsias

et al., 1999).

Another approach, multiple-image detection,

is based on the assumption that a fire-affscted

area will appear spectrally different on a post-

fire image compared to i t s appearance on an

image taken before the fire. Due to the large

changes caused by fire, methods that look for fire-

induced changes between dates ("change detection"

methods) usually detect burned areas more

accurately than single-date methods (Thompson

& Vink, 1 997; Hudak et al., 1998). For example,

urban areas and bare soils can appear similar to

burned areas on a single image, but will change

l i t t le between image dates, in discernible

contrast to most fire-induced changes.

Multiple-image methods, however, require

stringent preparation of imagery. Images must

be geographically registered accurately to one

another ("co-registered") to avoid "burned areas"

appearing between dates that are really just due

t o inaccurate registration. Co-registration of

images becomes less accurate with decreased

spatial resolution. lmages must also be radio-

metrically inter-comparable, i.e. the same band,

band combination or index from the same sensor

should be used for each image date.

Adjustments may also be necessary to nor-

malise the sensitivity of each image prior to their

comparison to try to prevent changes in viewing

geometry and atmospheric conditions between

dates from generating spurious changes in pixel

values that could be mistaken for burned are

(Viedma et al., 1997). Other disadvantages ar

that a minimum of two images per detect

halves the chance of obtaining a cloud-fre

I

(

(

r

L

C

1 1

t

C

S

s

tl

d

CI

fi

c;

bl

n i

(1

PC

Rr

CC

t i c

fe,

et

th

o t


and doubles the cost over single-image

techniques.

8.4.3.3 Other Considerations Relevant

to All Methods of Detection

Obscuration of burned areas

Smoke is relatively opaque at visible wavelengths,

can obscure burned areas at long-thermal infra-

red wavelengths, and has a small effect at NIR

wavelengths, all of which can complicate burned

area detection. However, at certain MIR wave-

lengths, even optically thick smoke plumes are

transparent (Miura et al., 1998). MIR-based

detection is therefore useful in areas where thick

smoke i s present for much of the burning

season, as is the case over much of Africa.

Thick cloud obscures the surface at visible to

thermal wavelengths and can confound remote

detection of burned areas. This is a particular

constraint when mapping large late dry-season

fires, which can be obscured by cloud. In such

cases, field mapping is sti l l necessary.

Dense tree canopies can "hide" fires that are

burning in the grass-shrub layer below, as was

noted in the Hluhluwe-Umfolozi Game Resewe

(Thompson, 1993).

Post-fire regrowth and greenup

Regrowth of vegetation following burning can also

confound detection. This can be a major limita-

tion in places where greening up begins within a

few days of burning (e.g. in Ivory Coast - Belward

et al., 1993), or if pre-cured grass burns early in

the season and has greened up before an image is

obtained. This affected the accuracy of mapping

of early season (pre-curing) fires in Pilanesberg

National Park, South Africa in 1 996 (Thompson &

Vink, 1997), with some small fires left detected

using a multi-temporal approach.

Green up poses less of a constraint in areas

where it is delayed until the onset of rains, as is

the case over much of Namibia, parts of South

Africa and Botswana.

Soil moisture

High soil moisture levels and consequently patchy

(low severity) fires can also confound detection in

certain circumstances, e.g. Pilanesberg National

Park in 1997 with late season rains (Thompson &

Vink, 1997). Wet soils can be much darker than dry

soils, and may be misclassified as burned areas.

Threshold variability

Variations in viewing atmospheric and surface

conditions at different places and times mean that

it is not usually possible to use the same fixed

numerical thresholds to classify pixels as burned

or unburned. Appropriate thresholds may be

chosen using field validation data (but these are

often lacking), by visual interpretation or using

statistically-based techniques. Visual determina-

tion is usually superior to statistical methods,

because it takes advantage of the superior pattern

recognition ability of the human mind. For example,

Salvador et al. (2000) attempted several objective

techniques for detecting burned area thresholds,

but found all to be inferior to visual assessment.

Interactive methods, however, require the analyst

to have a good knowledge of visual interpretation

of burned areas from multi-band imagery. Research

is ongoing to develop fully automated techniques,


but it is likely that visual checking of burned area

products will always be important.

Assess map accuracy using reference data

It is a good idea to check product accuracy by

gathering a representative sample of independent

reference data on burned and unburned areas,

with which to validate the burned area product.

Establishing map accuracy gives decision makers

confidence in using the remotely sensed products,

and can identify areas where the mapping method

needs further improvement. Congalton and Green

(1999) provide a review of the main methods

used to assess the accuracy of remotely sensed

data.

The upsides

Having discussed the pitfalls, it is important to

state some of the upsides of burned area mapping

using remote sensing. Existing semi-automated

methods (e.g. Flasse, 1999; Salvador et al., 2000),

if chosen and applied with care, can rapidly and

cheaply deliver products at sufficient accuracy for

fire management. In fact, since it is only required

to classify two classes (burned and unburned),

product accuracy should routinely exceed, for

example, the accuracy of remotely derived vegeta-

tion maps (since classification accuracy generally

increases as the number of classes decreases

[Sannier, 19991). Several studies have found

remote mapping of burned areas to be much more

accurate than ground-based mapping for capturing

the patchy nature of burned areas - including the

recording of unburned "islands" within larger

burns. For example, in the 48 000 ha Pilanesberg

National Park, Thompson and Vink ( 1 997) found

that field maps overestimated by 8500 ha (or

approximately 17%) the actual area burned,

resulting in an over-estimate of 39.5% compared

with satellite-derived burned area maps. Section

8.7 will give example of use of burned area

products in operational activities.

8.4.4 Vegetation Monitoring

8.4.4.1 Vegetation Products in Fire

Management

Vegetation monitoring provides important infor-

mation for understanding fire behaviour, includ-

ing ignition, growth and rate of spread (Cheney &

Sullivan, 1997), and is therefore crucial to help

land managers optimise both fire prevention and

fighting activity. Preventive actions in the USA,

Europe, Africa and Australia include the use of

prescribed fires.

In grassland and savanna with seasonal drought,

fires during the dry season are limited by grass

fuel availability, and grass productivity is in turn a

function of soil moisture availability from the pre-

ceding rainy season (Scholes & Walker, 1993).

Thus, fire frequency declines as precipitation

declines through an indirect yet strong relation-

ship. In forests, fuels accumulate over dekadal

time scales, and fire frequencies are much lower,

with fires occurring during episodic droughts. In

grassland, savanna or forests, fire frequency and

intensity depend on ignition sources, fuel charac-

teristics ( e g distribution, compaction, types,

moisture content, accumulation and flammability

[see Trollope, 1992]), and the vegetation land-

scape mosaic (Christensen, 1 98 1 ). Shifts in fire

frequency lead to changes in vegetation structure,

K$Ji:

2rf d

acd t

diffic

F 1

field

local

sensc

regic

usef~

Seve~

O b s ~

tar v

Elect


\idhich in turn modify the intensity of subsequent

fires (Kilgore, 198 1 ).

The important vegetation characteristics to

he taken into account in fire management are

therefore: Fuel load (influencing fire intensity),

noisture content (influencing both fire ignition

2nd spread), continuity (influencing fire spread)

and height (influencing height of flames and hence

difficulty of suppression).

Fuel characteristics may be measured in the

field, but such measurements only represent

local conditions at a few locations. Remotely

sensed data provide information at landscape,

regional and global scales, and are therefore more

useful for land managers.

8.4.4.2 Vegetation Monitoring Systems

Several different sensors currently on board Earth

Observation System satellites are used to moni-

tor vegetation in three different portions of the

Electromagnetic (EM) spectrum.

Visible to shortwave infrared (0.40-2.50 mm,

previously defined also as visible, NIR, SMlR

and LMIR). Vegetation reflectance in this por-

tion of the spectrum provides information on

vegetation biophysical parameters such as

chlorophyll, physiological structure and leaf

cellular water content (Tucker, 1980). Chloro-

phyll absorbs the red and blue elements of the

EM spectrum, internal leaf structure makes

vegetation highly reflective in the near-infra-

red and leaf cellular water absorbs radiation

in the shortwave infrared. Satellite band

combinations of different regions of the EM

spectrum (also called vegetation indices)

emphasise the spectral contrast between the

different regions of the EM spectrum and

allow hidden information to be retrieved.

Vegetation indices are empirical formulae

designed to produce quantitative measures,

which often relate to vegetation biomass and

condition (Gibson & Power, 2000; Verstraete

& Pinty, 1996). The most commonly used

vegetation index is the Normalised Difference

Vegetation Index (NDVI):

(NIR - red) NDVi =

(NIR + red)

where NIR is the reflectance measured in the

near infrared channel and red the reflectance

measured in the red channel; the higher the

NDVI value, the denser or healthier the green

vegetation. Visible and near-infrared channels

are available on most optical satellite sensors

including NOAA-AVHRR, EOS-MODIS, SPOT-

VEGETATION, SPOT-HRVIS, LANDSAT-TM,

and LANDSAT-MSS. Other indices, such

as the SAVI, TSAVI, ARVI, GEM1 (see Flasse

& Verstraete, 1994, for more details), have

been developed to identify the presence of

vegetation and to be less affected by perturbing

factors, such as soil colour and atmospheric

contamination.

To advance further the performance of

such spectral indices, a method has now been

proposed by Verstraete and Pinty ( 1 996) to

create an optimised index for specific sensor

characteristics. In any case, it is important

that users carefully choose the appropriate

index to best respond to the requirement of

their work. tidar is an active remote sensing


system based on laser altimetry principles

that operates in the near-infrared portion of

the spectrum, where green vegetation is

highly reflective. Lidar accurately measures.

tree heights and has been used to estimate

forest canopy volume, which has been shown

to be a good indicator of biomass and leaf area

in high biomass forests of the US Pacific

Northwest (Lefsky et al., 1 999b). No satellite

lidar systems have yet been launched, but the

Vegetation Canopy Lidar (VCL) satellite is cur-

rently being constructed.

Thermal infrared (6.0- 15.0 mm). Emittance

in this portion of the EM spectrum provides

information on the thermal properties of

vegetation cover, such as sensible heat. Heat

measured by satellite sensors is used to esti-

mate evapotranspiration of vegetation canopies,

which can be a good indicator of water stress

(Moran et al., 1994). Thermal infrared bands

are available on sensors such as NOAA-

AVHRR, METEOSAT, and LANDSAT-TM.

Microwave (0.1-1 00 cm). Active and passive

microwave approaches have been developed

to sense soil water content, which can be

highly relevant to vegetation monitoring (Du

et al., 2000). Passive microwave sensors pro-

vide information on the thermal properties

of water (Schmugge, 1978). Passive sensor

SSMII is currently available on the Defense

Meteorological Satellite Program (DMSP)

platform. Active microwave sensors provide

information on the dielectric constant, which

may be related to vegetation water content

(Moghaddam & Saatchi, 1999). Active sensors

currently available include RADARSAT and

ERS-2, and ENVISAT-ASAR from October

200 1 .

8.4.4.3 Operational Vegetation Products

The main vegetation products useful t o fire

management are:

Fuel load. Estimation of biomass is performed

using optical sensors. Biomass maps were

derived in the grassland regions of Etosha

National Park, Namibia, using NDVl com-

puted from NOAA-AVHRR images (Sannier

et al., 2002). Similarly, Rasmussen (1 998)

estimated net primary production in Senegal.

However, these studies are spatially limited

and more work is required on refining the

relationship between biomass and the NDVI

for different vegetation communities. Lidar

data may one day prove useful for measuring

and monitoring forest biomass, but are still

mostly unavailable.

Vegetation moisture content . Operational

estimation of vegetation water content i s

performed using optical and thermal infrared

sensors. The use of radar sensors to monitor

vegetation water content requires further

research before it will be operational (e.g.

Moghaddam & Saatchi, 1999).

Three methods are used to estimate vege-

tation water content. The first method uses

the Normalised Difference Vegetation lndex

(NDVI) to estimate live vegetation chloro-

phyll and moisture content (Burgan, 1996).

The NDVl is used t o compute a Relative

Greenness lndex (RGI), which is incorporated

with weather data to define a Fire Potential


lndex (FPI) (Burgan et al., 1998). The FPI is

computed for assessing forest fire hazards in

the Mediterranean climate region of southern

California (USA) (http://edcsnw3.cr.usgs.gov/

ip/firefeature/firepaper. htm).

Similarly, the Fire Potential lndex has been

adopted by the Natural Hazards project of

the Space Application Institute, joint Research

Centre (Ispra, Italy) to evaluate forest fire

risks in Europe (http://natural-hazards.aris.

sai.jrc.it/fires/risk/). However, Ceccato et al.

(200 1 a) recently showed that the relationship

between degree of curing and vegetation

moisture content is not applicable to all types

of vegetation.

The second method estimates the moisture

content through the measurement of

evapotranspiration, an indicator of vegetation

condition. Evapotranspiration, as measured

by thermal sensors, may be estimated with

several indices: the Crop-Water Stress lndex

(CWSI) (Jackson et al., 198 I), the Stress

lndex (SI) (Vidal et al., 1994), and the Water

Deficit lndex (WDI) (Moran et al., 1994).

However, it has been shown that many

species may reduce evapotranspiration with-

out experiencing a reduction of water content

(Ceccato et al., 200 1 a).

The third method is based on direct measure-

ment of vegetation water content and uses

the absorption property of water in the

shortwave infrared (spectrum between

I. I3 mm and 2.50 mm). Using a combina-

tion of the shortwave infrared and infrared

wavelengths from SPOT-VEGETATION, a

. Global Vegetation Moisture lndex has been

created to measure directly vegetation water

content (Ceccato et al., 2002a). This method

is currently being tested for fire management

applications (Ceccato et al., 2002b).

Vegetation continuity and density. High-reso-

lution satellites are needed to characterise

the spatial structure of the vegetation canopy.

Hudak and Wessman (200 1) have shown that

a textural index of high-resolution imagery

serves as an accurate indicator of woody plant

density in semi-arid savanna.

Vegetation height. Estimation of vegetation height

is still at a research stage. Synthetic Aperture

Radar (SAR) studies are being developed to

estimate vegetation height (Sarabandi, 1 997),

but are not yet operational. Lidar provides

direct, accurate measurements of canopy

height but are currently limited in spatial

extent and availability (Lefsky et al., 1999a).

8.4.5 Rainfall Estimation

8.4.5.1 Derivation of Rainfall Estimates

Rainfall is normally measured using rain gauges.

However, the network of rain gauges may be

sparse in those areas affected by fire. Satellite

observations are used in combination with, and

to augment, rain gauge data. Satellite data pro-

vide a spatially complete, uniformly distributed

coverage that allows better estimation of rainfall

where rain gauges are infrequently and irregu-

larly sited.

Meteorological satellites in geo-stationary

orbit (i.e. an orbit where the satellite appears fixed

at the same point in the sky) are able to collect

images of a large area frequently. For example,


the Meteosat satellite collects an image of the

whole of Africa and Europe every 30 minutes.

Similar satellites are available for other parts of

the Earth. The frequency of images collected by

these satellites is important, as i t allows rain

clouds to be located and tracked, which is vital

data for producing accurate rainfall estimations.

Data from polar orbiting satellites (which move

across the sky) can be used to estimate rainfall

(e.g. using passive microwave data) but these data

are available much less frequently for each loca-

tion on the ground.

Geo-stationary satellite rainfall measurements

are particularly appropriate for areas where rain-

fall comes mainly from convective clouds. Con-

vective clouds are formed when small warm

lumps of air (called thermals) rise up to produce

clouds. On meteorological satellite images these

clouds appear firstly as small and round and, as

they grow, become colder. They cool down as

they rise and thicken into storm clouds. The tem-

perature of the cloud top can easily be measured

(both day and night) using the thermal infrared

waveband data. It is possible to predict whether

particular clouds will produce rainfall because the

colder (and thicker) the cloud, the more likely it

is that rain will fall. The duration of the cold cloud

in any particular location can be measured quite

precisely as images are available so frequently. A

simple linear relationship between cold-cloud-

duration (CCD) and the amount of rain produced

is used as the basis for a first indication of the

quantity of rainfall. Local rain gauge data is used

to calibrate the rainfall estimation for each loca-

tion. This technique is called the cold cloud pre-

cipitation method.

The cold cloud precipitation method does not

work so well for estimating rainfall from other

types of cloud. For example, layer (stratiform)

clouds form when air rises consistently, eithei-

by night-time cooling or by clouds associated with

weather fronts. For these types of clouds the

relationship with rainfall is more complicated.

Hence, cold cloud precipitation method of rain-

fall estimation works well in the tropics where

most of the cloud is convective, but less well in

mid and high latitudes where other types of cloud

are dominant.

The rainfall estimations are usually built up

over period of approximately ten days, called a

"dekad" (this is a standard reporting period for

meteorological data). There are three dekads in

each calendar month. The first dekad of each

month begins on the I "; the second dekad begins

on the I I '" and the third dekad begins on the 2 1 %

(and hence will vary in length depending on the

particular month). Figure 8.10 provides an

illustration of dekadal CCD.

8.4.5.2 Sources of Rainfall Estimates

The TAMSAT (Tropical Applications of Meteorol-

ogy using SATellite and other data) group, at the

University of Reading (UK) researches the use of

satellite imagery for estimating rainfall. More

details of the cold cloud precipitation method and

how it can be applied is given in Milford et al.

( 1 996). They have produced a Rainfall Estimation

Workbook (Grimes et al., 1998) to introduce prac-

tical rainfall estimation techniques. An extensive

l ist of publications and the latest on-line dekadal

rainfall estimate is provided on their website at

http://www.met.rdg.ac.uk/tamsat/.

Th

Devel

Syster

lated

effort:

is base

but in

deal b

Arkin,

use nu

tive h~

the cot

ferent

ture of

rainfall

FEV ing dail*

able foi

Disserr

/edcintl

AD[

satellite

Figure


Southern African CCO i w agricultural statistics and digital map data. Ram-

fall charts can be viewed on-line or downloaded.

Software to store, analyse and display rainfall data

can also be downloaded.

The United States Agency for International

E.-suelopment (USAID) Famine Early Warning

S;>stem (FEWS) produces estimates of accumu-

Iriad rainfall to assist in drought monitoring

e!.ii>iTs in sub-Saharan Africa. Their methodology

is ixsed on the cold cloud precipitation method,

biit incorporates some other data where they

deaf better with non-convective rainfall (Xie &

Ar-kin, 1997; Herman et al., web publication). They

uz3 numerical models to produce wind and rela-

ti !? humidity information and take into account

ti-::: contribution of orographic rainfall (where dif- r .. ~ s ! cnt rainfall patterns are produced by the struc-

twe of the terrain). Satellite-passive microwave

r&k4I estimations are also incorporated.

FEWS rainfall estimation (RFE) data, includ-

ing daily, dekadal and historical archives, are avail-

aljie for sub-Saharan Africa from the Africa Data

Wikmnination Service (ADDS) website at http:/

/edcintl.cr.usgs.gov/adds/adds.html.

ADDS also hosts other data sets, including

smdlite-derived dekadal vegetation information,

8.4.5.3 Rainfall Data in

Fire Information Systems

Rainfall estimations can be produced for finely

gridded areas, e.g. 5 x 5 km areas, but are often

used as summaries over political or physical

regions, for example, countries or catchments.

Statistics can include total, mean and standard

deviation of rainfall in millimetres, area of rainfall

coverage within a region, etc. The rainfall data can

be combined with other data to produce further

information, for example, hydrological modelling

or a fire information system.

Rainfall maps can be used to inform manage-

ment, for example, recent rainfall could to be

taken into account when deciding the timing of

prescribed burns (Carlson, 200 1 ).

Rainfall estimation can also be integrated with

other data, for example, as an input to fire risk

assessment (Aguado et al., 2001). The rainfall

information i s incorporated in estimation of

vegetation moisture content, which can then be

combined with fuel data for assessing fire risk.

8.5 IMPLEMENTING REMOTE SENSING

IN A FlRE MANAGEMENT CONTEXT

This section looks briefly at some of the resources

typically required to run a small remote sensing

component to contribute to fire management.

Data produced by any remote sensing activities

should be integrated into a fire management infor-

mation system so that, through the combination

F!i_;ure 8.10. Example of a Dekadal Cold Cloud Duration (CCD) image (one slot equates to 30 minutes).


of data from various sources, more information

can be extracted to better support management

decision-making.

A range of situations could occur and some

fire management teams may have access to their

own remote sensing group or government-run

remote sensing resources. Many others may find

some existing local expertise in remote sensing,

e.g. a local consultancy, scientific institute or

university who could assist in the setting up of a

remote sensing group, provide training or even

be contracted to do the work. Alternatively, an

in-house specialism could be developed.

It is important to have a general idea of what

is wanted out of any remote sensing endeavour

so that sensible levels of resources can then be

allocated to achieve the desired outputs. A remote

sensing expert should work with management to

identify areas where remote sensing can contrib-

ute and help design an overall remote sensing

strategy. The following are some considerations

for those who want to set up and run a remote

sensing component for fire monitoring and manage-

ment.

8.5.1 Skilled Personnel

The person running the component should have a

combination of remote sensing skills and field

experience (perhaps in fire or vegetation ecol-

ogy), or at least demonstrated aptitude and a

willingness to learn new skills. He or she should

have some input to design of a remote sensing

strategy, decide which imagery will be used to

obtain the information, and how often, and choose

and implement the methods to extract informa-

tion. The person should make every effort to

assure quality control at all stages, including

assessing the accuracy of final products wherever

possible.

Once procedures are in place for prepzing

operational products, there will be routine image

processing and data input tasks to complete. These

tasks are essential to maintain a fire manT:ge-

ment information system and allow use ol' the

latest information.

It is also important to have a person arc?;~nd

who completely understands how the informa-

tion system operates and can draw awareness to

and help users to exploit the potential of ehe

system. These functions are best sewed in-house.

Hence, the most appropriate person may be

someone with local knowledge of fire conditions,

and the required outputs of a fire management

information system. In most cases, computer

literate staff can also be trained to maintain an

information system.

8.5.2 Access to Relevant lnforrnation

Staff should ideally have access to publications on

remote sensing, land mapping, fire ecology and

other relevant information. Easy access to up-to-

date literature is especially helpful in seleciing

appropriate methods to deliver particular informa-

tion products and in avoiding common mistakes.

Collaborative research with local and international

scientific institutions can also help in product

development.

8.5.3 lnfiastructure

Office space is required for in-house remote

sensing, and adequate remote sensing hardware

and software must be acquired. Image processing


2nd Geographic lnformation System (GIs) tools

2iEow a fire information system to be built up

rpj i th the objective of supporting operational fire

I-ixmagement.

Computer hardware will typically include at

least one PC with a high-speed processor, large

hard disk and lots of memory. A large monitor

is also useful, allowing images to be seen in rea-

sonable detail without the need for excessive

zooming.

There will also be data archiving and retrieval

capacity, e.g. CD readlwriters are currently a very

&tractive option, since most receiving stations

now provide remotely sensed images on CD, due

to the low costs for this media. A colour printer

is essential for the presentation of the various

products, while a Global Positioning System

(GPS) receiver can be very useful in collecting

accurately located field data for plotting on geo-

referenced images. A digitising table is also useful

so that information held on topographic and other

p p e r maps can be integrated with remotely

sensed images and other data in a GIs.

Image processing software provides the tools

required to pre-process imagery into usable form

(e.g. to correct and calibrate raw imagery and

transform it to a standard map projection) and to

develop and apply algorithms to process data into

information products. They also have sophisti-

cated tools for investigating, displaying and

enhancing the appearance of imagery.

GIs software is also required for a sophisti-

cated information system that allows integrated

analysis of many layers of spatially registered

information. GIs tools can allow remotely sensed

products to be analysed in the context of any other

geo-located information held by management,

such as maps of infrastructure, administrative

boundaries, planned ignition points, fire history

and maps made in the field (e.g. vegetation type,

fire severity, etc.).

Spatial models can be built up through the

arithmetical combination of information in the

different layers. For example, fire danger might

be estimated by combining remotely sensed in-

dicators of vegetation state and standing biomass

with maps of roads and population centres and

synoptic meteorological data.

8.5.3.1 Adequate Budget to Maintain

the Information System

As well as personnel and initial set-up costs for

hardware and software, the budget should include

allocations for recurrent expenses, such as

image data costs and additional data acquisition.

Maintenance and upgrade costs for hardware and

software, and replacement of consumables, must be

considered too. Fieldwork is necessary to validate

remote sensing outputs, so a budget for transport

(and maybe equipment) should be available. It

should be remembered that, for monitoring pur-

poses, it is important to maintain the scale of data

and the frequency of acquisition, so that quality and

consistency of information products are maintained.

Hence, budget constraints are very important in

developing an operational remote sensing strategy.


8.6 EXAMPLES OF APPLICATION OF

REMOTE SENSING PRODUCTS IN

FIRE MANAGEMENT IN AFRICA

8.6.1 South Africa

The use of remote sensing technologies for fire

management in South Africa can be divided into

three basic application areas - namely, post-event

burned area mapping (including associated fire

intensity / severity analysis), active fire monitor-

ing, and biomass estimation (in terms of fuel loads

and potential fire risk assessment) - and are

primarily concerned with the fire-prone savanna,

grassland and fynbos biomes.

Post-event burned area mapping is arguably

the most common application area, and has

generally been conducted as a parallel research

orientated activity in support of more operational,

traditional field-based mapping. Many of the larger

protected areas in the savanna biome have tested

this kind of image-based fire mapping with a fair

degree of success, e.g. Kruger National Park

(Hetherington, 1997; 1998), Pilanesberg National

Park (Thompson & Vink, 1997), Madikwe Game

Reserve (Hudak et al., 1998), Mkuze Game

Reserve and Hluhluwe-Umfolozi Game Reserve

(Thompson, 1990; 1993).

Of key significance in many of these projects

has been the obvious improvement in both the

accuracy of individual burn area delineation, and

the identification of small isolated non-burned

"islands" that are often missed during more

generalised field-mapping. For example, field-

mapped estimates of total fire extent differed

from image derived estimates by 50.4 %, equiva-

lent t o 4252 ha (4.5 % of total reserve area) in a

study completed by Thompson (1 990; 1993) in

Hluhluwe-Umfolozi Game Reserve (although it

was noted at the time that under tree canopy fire

scar extents were difficult t o define on the

imagery). Similar results have been reported far

studies in Pilanesberg National Park, where field

maps over-estimated the total burned area by

8500 ha (or 17% of the total reserve area)

(Thompson & Vink, 1977). In this case field esti-

mates were 39.5% higher than the satellite-

derived estimate. In general, these historical fire-

mapping exercises have used high resolution

Landsat or SPOT multispectral imagery, for de-

tailed mapping at scales in the order of 1 :SO 000

to l:75 000.

Image classification problems tend to arise, as

would be expected, when the image acquisition

date is significantly different from the burn event

date, especially if post-fire regrowth or green-up

of the vegetation has occurred in the interim

period. Additional classification problems can also

be experienced if the prevailing environmental

conditions at the time of the burn did not result

in a clean burn with a clearly definable extent.

The compilation of end-of-fire season fire scar

maps for the Pilanesberg National Park for the past

several years has indicated that no single image

processing technique or algorithm is optimal for

all conditions (especially if it is necessary to use

sub-optimal imagery in terms of acquisition date

in relation to actual fire event or precipitation

patterns). Rather a range of data processing

techniques are necessary to cover all possible

conditions. For example, post-event fire scar

mapping for the years 1994 to 200 1 in Pilanesberg

has involved the use of both single and multi-date


ii-wgery, derived indices, simple-level slicing,

!r:c data clustering models, and principal component

zm~ysis to map fire scars. In most cases, this has

keen based on Landsat Thematic Mapper imagery

(:.zfids three, four, five and seven), or closest

!. ? 37- equivalents.

Recent work used Landsat images to establish

2 ,ire history for Madikwe Game Reserve and

zcwounding farms, including Botswana's Kgatleng

area and southern district (Hudak & Brockett, in

press). Fire history was derived from burned

areas mapped from 22 annual fire maps from the

period 1972 to 2002 (excluding 1974, 1975 to

1978, and 198 1 to 1985).

Research has been conducted in terms of fire

severity and intensity mapping using near-real

time imagery, linked to internal fire scar charac-

teristics. A key area of activity being studies linked

Figure 8.6 1. NASA ER-2 aircraft image of a prescribed SAFARI fire over theTimbavati Reserve. Higher confidences on the

position of the flaming front and fire emission factors can be determined from ER-2 MODIS simulator data at a resolution of

50 In. Fire ground variables such as flame height, climate parameters and rate of spread are measured coinciding with TERRA a d ER-2 overpasses.


to South Africa's contribution to the international

SAFARI 2000 initiative. This has primarily involved

the assessment of products derived from EOS-

MODIS*, within the context of park fire manage-

ment activities and the development of automated

fire monitoring systems. Case studies are cur-

rently being conducted in both Kruger National

Park and Madikwe Game Reserve, where to date

over 80 fuel measurements involving prescribed

burns have been recorded. The MODlS fire algo-

rithms can be transitioned into an operational

monitoring system to render accurate and timely

information on the location, spatial distribution,

intensity and timing of fires in South African con-

servation areas. These findings could support park

management objectives that monitor and modify

long-term fire management programmes in rel-

evance to fire regimes.

This study involves validating MODlS burn scar

data on a near-real time basis using co-located

Landsat 7 ETM 30 m resolution fire maps, com-

bined with field data on combustion intensity and

completeness. Within the 2000 SAFARI field cam-

paign, grey scale ash colour, biomass observa-

tions and field spectrometer recordings (using a

hand held ASD radiometer) of burn scars were

sampled, since ash colour is postulated as being a

retrospective measurement of fire intensity

(Stronach & McNaughton, 1989).

Post-fire burned area mapping in the fynbos

biome is somewhat different from that in the

savanna (or grassland) biome, since individual fire

scars in fjmbos can remain visible for many seasons

after the actual fire event due to the slow regen-

eration of the local vegetation. Such conditions

can make the temporal separation of inter- and

intra-year fire scars problematic, although

Thompson ( 1 990; 1993) reported that the use of

within-scar NDVl difference was successful for

age and sequence determination of historical fire

scars in this area.

Satellite imagery received a major boost locally

as an operational tool during the December 1999

- January 2000 wildfires in the Western Cape,

which burnt vast tracts of mountain and coastal

fynbos communities on the Cape Peninsular and

West Coast. During this period a series of SPOT

images was specially acquired to provide near-

daily coverage of the fires and their rate of spread

in, often inaccessible, mountains. Although this

information was not used for true real-time fire

management activities, it proved a useful tool for

public-level media instruction as well as for

post-event, disaster management assistance and

planning.

Biomass monitoring is a key component of pre-

fire risk assessment. Several case study examples

illustrate the potential of remote sensing for this

application, although, as with post-fire mapping,

these are primarily research rather than opera-

tional level studies.

Studies in both the Hluhulwe-Umfolozi Game

Reserve and Drakensberg mountains have indi-

cated that pre-fire season predictions of potential

fuel loads (tons 1 ha) can be achieved with a high

degree of accuracy (i.e. r2 D 0.8) in both savanna

and grassland areas using Landsat and SPOT

equivalent data. These are typically based end-of

growing season NDVI-based biomass models,

which have been calibrated with actual field-

derived biomass data (Thompson, 1990; 1993;

Everson & Thompson, 1993).

* In February 2000 Terra-AM, the flagship platform of NASA's Earth Observing System (EOS), began collecting what will

ultimately become part of a new 18 year data set (Kaufman e t a/. 1998). The MODerate resolution Imaging Spectroradiometer

(MODIS) onboard TERRA senses the earth's surface in 36 spectral bands and can provide daily coverage of South Africa at a

nadir resolution of 250 m and 500 m in the visible to near-IR and I km resolution in the thermal spectral range.

I*lore recently a similar research project

losked at using coarser resolution I x I km SPOT

VEGETATION NDVl products for determining

end-of growing season fire risk along power trans-

mission lines on a national basis, since wildfire

esrnbustion effects can result in transmission

interrupts in some instances (Thompson & Vink,

206 1).

Near-real time monitoring of local fire events

is starting to become established as a viable tech-

nique at both local and national scales. For example,

since November 2000 the ARC-ISCW* has been

part of the World Fire Web (WFW), participating

as the Southern Africa node. WFW is a global

network of computers that detect active fires using

daily NOAA-AVHRR satellite data. The network

is co-ordinated by the European Union's Joint

Research Centre (EU-JRC) in Ispra, Italy. The

input data are daily NOAA- 14 AVHRR afternoon

passes (Ahern et al. 2000). Daily fire maps are

ccmpiled at each regional node and then made

available in near real-time on the World Wide

W b (WWW)**. The fire information can be

downloaded in a text format giving latitude and

longitude of each I km AVHRR pixel detected

that contains a fire on that day. improvements to

the software will enable NOAA- 16 imagery t o be

processed.

8.6.2 Namibia

8.6.2.1 Introduction

Namibia lies in the west of Southern Africa, bor-

dering Botswana and South Africa t o the east and

south, the Atlantic Ocean to the west, and cover-

ing an area of approximately 824 000 km2.


Large areas burn each year. Almost five million

hectares burned in 2000, whilst in years of lower

rainfall this figure is significantly lower (Le Roux,

200 I). Excessive indiscriminate burning is having

highly negative effects on some ecosystems,

whilst in other areas, fire frequencies are more

in equilibrium with requirements for the long-

term stability of existing vegetation communities

(Goldammer, 1998).

Fires burn during Namibia's severe dry season

from April to October, mainly as surface fires

that spread in the grass and shrub layer. Crown

and ground fires occur over only limited geo-

graphical areas. The amount and connectivity of

surface fuel i s highly variable spatially and tempo-

rally, controlled by a severe rainfall gradient

orientated in an approximately SW to NE direction.

The most frequent, intense and extensive fires

occur in the north and northeast, whilst fires

occur infrequently in the south and west. Figure

8.12 shows a burned area map of the fire-prone

areas of Namibia (derived from remote sensing)

and demonstrates the general increase in the size

and extent of burned areas from SW to NE. Light-

ning ignited fire is the most significant natural

cause, but accounts for only a small percentage of

all fires. The majority of fires are anthropogenic,

either set deliberately or accidentally (Goldammer,

1 998).

8.6.2.2 Fire Management Issues

in Namibia

Figure 8.6.2b shows the six fire regime zones of

Namibia (Trigg & Le Roux, 200 I), as a framework

for describing fire management issues in Namibia.

In zones I and 2, low rainfall means that fires

* Agricultural Research Council: Institute for Soil, Climate and Water. ARC-ISCW (Pretoria), houses a fully calibrated NOAA- AVHRR I km data set for Africa south of the equator with data from July 1984. Such an archive is ideally suited to the

development of long-term fire frequency models for the region, and can contribute to the sustainable management of

ecosystems as well as for global carbon management.

" '' See http://ptah.gvm.sai.jrc.it/wfw/ or http:/lwww.arc-lscw.agric.za/main/fireweb/index.htm

I 89


occur only rarely, have low intensity and relatively

insignificant impacts. In zones 3 and 5, widespread

pastoralism means that fires are generally not

desired because they result in a loss of forage.

Fires are suppressed by farmers' associations in

zone 3 whenever possible, and occasionally by

local communities supported by the Ministry of

Agriculture in zone 5.

In the Etosha National Park (zone 4), fire is

managed by the Directorate of Resource Manage-

ment (DRM) of the Ministry of Environment and

Tourism (MET), using a park block burning pro-

gramme intended to maintain or improve bio-

diversity (Stander et al., 1993; Du Plessis, 199Pj.

In areas of Kavango and Caprivi in the east ~f

zone 6, very frequent fires pose a serious threat

to large areas of wooded and forested lalid

(Mendelsohn & Roberts, 1997). In East Caprivi,

communities were mobilised by the Directorzte

of Forestry (MET), with support from FINNIDA,

to clear fire lines to retard fire spread and fre-

quency in wooded and forested areas. In some

areas, fire plays a more positive role, for instance

in the regeneration of grasses used for thatching.

Guidelines on burning have been prepared that

recognise the need burn in some areas and to

exclude fire in others (Trollope & Trollope, 1999).

8.6.2.3 Fire information and Management

Fire statistics are not yet compiled or aggregated

at a national level, and resources for obtaining

them in the field are limited. The most compre-

hensive surveys of active fires and areas burned

have been made using image data from the

Advanced Very High Resolution Radiometei-

(AVHRR) sensor onboard the US NOAA (National

Oceanic and Atmospheric Administration) satel-

lite series, as indicated in Cracknell (1997). The

AVHRR data are provided by a PC-based receiver;

installed at the Etosha Ecological Institute (EEB-

within zone 4) and run by Park management.

8.6.2.4 Remotely Sensed Products

and Their Uses

Policy level

AVHRR-based maps showing the distribution and

approximate timing of burned areas over the fire-

Figure 8.1 2. Fires in northern Namibia, for the 1997 burning season, colour coded according to approximate date of burn.

Figure 8.13. Six major fire regime zones of Namibia.


pxne areas of Namibia (e.g. Figure 8.12) have

b2.w incorporated into environmental profiles of

tks Caprivi and North Central regions. These

&?uments are designed to place environmental

inbrmation into the hands of politicians and other

dxision-makers (Mendelsohn & Roberts, 1997;

i+ndelsohn et al., 2000). AVHRR-based maps have

ziso been included in the latest fire policy docu-

rnmt for Namibia (Goldammer, 1999), to show

ti-:r:; distribution of burning. AVHRR data were also

il-xsgrated with maps of vegetation and land use

ti:: stratify Namibia into different fire regime

ZCacs.

Fire frequency was then estimated from

A~!i-lRR imagery for the three most fire-prone

fii-t? regime zones (Trigg, 1997; Le Roux, 200 1).

Fiprres 8.14 and 8.1 5 show that fire frequency is

n-ruch higher in zone 6 (Kavango and Caprivi) than

elsewhere in the country, with the majority of

Ceprivi burning two to four times in just a four-

year period (compare Figure 8.15[a] and Figure

8. i 5 [b]). Extensive areas of wooded and forested

land in the Caprivi and Kavango are particularly

~!i?der threat from such frequent fires (Jurvelius,

pe:-s.com., 1 998). These remote studies help to

identify areas where fire frequency is too high for

the intended land use,'and for refining fire policy

and management strategies.

Management level

Etosha Ecological Institute (EEI)

Active fires are usually visually interpreted from

AVHRR channel 3 imagery within minutes of the

satellite passing overhead, and their centre

coordinates noted. This method has identified

several undesired fires prior to their detection

by management staff in the field, for more timely

response.

Implementation of a block-burning programme

requires mapping of all areas that burn within

the park each year. During the 16 years of pre-

scribed burning prior to the routine availability of

AVHRR-based burned area products, this was

done by either driving block perimeters or by

sketching extents onto a base map from airborne

observations. These methods were regarded as

reasonably accurate for large, cleanly burned

blocks, but inaccurate for heterogeneously burned

blocks containing large islands of unburned veg-

etation. These methods of mapping burned areas

also required a lot of time and resources. Since

1996, AVHRR data have enabled burned areas to

Figure 8.14. The number of times the areas of zone 6 (routinely monitored by AVHRR) burned over a four-year period (I??&-1999).


be mapped, in many cases more accurately and

with significantly less expenditure compared to

field-based methods. Using a simple change

detection technique, it now takes about two

hours per month to map burned areas over the

park and adjacent areas.

To plan new ignitions requires standing

biomass to be estimated for each burn block, in-

formation that is very difficult to obtain for large

areas using field-based methods alone. An

AVHRR-derived image of the maximum NDVl

attained at each pixel location during the previous

growing season is used to identify candidate blocks

with sufficient biomass for burning during the

next dry season. The biomass of one or more

candidate block is then surveyed using a disc

pasture meter to assist the final selection.

Caprivi

AS part of their evaluation of the success of ~ ( 7 ~

community-based fire control project, the Dii-ec-

torate of Forestry (DoF) wanted to assess i:he

impact of fire lines (cut by local communities) in

reducing the annual area of burn in East Capri'>ri.

An AVHRR-based study was commissioned and

found that less of the area had burned in the yea-s

since the onset of fire-line construction, suppor-t-

ing the assertion that the fire control efforts had

been worthwhile (Trigg, 1997). Burned area maps

were also used in public awareness campaigns

designed to educate people about the detrimen-

tal effects of very frequent fire.

8.6.2.5 Research and Development

Research at EEI and in Caprivi (Trigg & Flasse,

2000; 200 I), helped to quantify and improve the

accuracy of the remote sensing of burned areas in

Namibia. These studies used field and Landsat

TM- based su weys to assemble accurate reference

data on burned areas. The various reference data

were then used to assess the accuracy of existing

burned area products from AVHRR and to

develop new algorithms to detect burned areas

using data from other sensors such as SPOT

VEGETATION and MODIS (launched December-

1 999).

8.6.3 Botswana

Botswana's terrain consists almost entirely of a

broad, flat, arid subtropical plateau, though there

are hills in the eastern part of the country. In the

north-west, the Okavango River runs into the sands

of the Kalahari. The Chobe National Park is a beau-

tiful grassland reserve, popular for its large elephant

Figure 8.1 5. Percentage of land that has burned a different numbers of times within a set number of years: (a) shows the

percentage of zone 5 that burned between 0 and five times during a five-year period ( 1 994-1 998); (b) shows the percentage

of zone 6 (east of 2 I "E) that burned between 0 and four times during a four-year period (1 996-1 999).

ppdation. The Kalahari Desert, a varied envi-

rc31;inent of sand, savanna and grassland, covers

lzige parts of the country. This is an important

wildlife area, in which Botswana's two largest

coisservation areas, the Central Kalahari Game

Reserve and the Kgalagadi Transfrontier Park (in-

clding the Gemsbok National Park in Botswana

and the Kalahari Gemsbok National Park in South

Aii-ica), are located. The country has a sub-tropical

climate arid in the south-west. Maximum daily

temperatures vary from 23-32" C. During the

winter months, May to September, there are

occasional overnight frosts. The majority of the

rain falls in the north and the east, almost all in

summer between October and April.

8.6.3.1 The Fire Issues in Botswana

Fire is a natural occurrence in Botswana. The

vegetation types and ecosystems across most of

the country have evolved with fire as a major

shaping force.

With the correct frequency, timing and extent

of burning, fires can have many positive effects

inchuding maximising range productivity, promoting

species diversity and controlling bush encroach-

ment. Many rangeland vegetation types actually

require a combination of fires, grazing or browsing

to maintain the species diversity and productiv-

i ty.

However, too often, fire seems to be used

inappropriately, or to get out of control, resulting

in undesired effects. Fire monitoring has shown

that large areas of Botswana burn every year with

an average of 8-15% (48-90 000 km2) of the

country affected each year. These large and fre-

quent wildfires cause damage to property and


threaten lives, damage forest, reduce grazing avail-

ability, change vegetation ecology, increase the

impact of drought, increase soil erosion and land

degradation and cause increased wildlife and live-

stock mortality.

The management and control of bush fires in

Botswana is a critical issue for the sustainable

development of the livestock, forestry and wild-

life sectors. Fire is already extensively used as a

range management tool throughout the southern

African region for maximising rangeland productiv-

ity. The Ministry of Agriculture and other govern-

ment departments have recently put fire manage-

ment on their agenda as a priority and are devel-

oping policies to improve knowledge on the

potential problem and strategies to tackle the

issues. It is recognised that many issues need to

be addressed, such as improving knowledge on

fire and wildland ecology, knowledge on the socio-

economic causes and consequences of fires, and

community education on fire management.

8.6.3.2 Fire Information and Management

In Botswana, the Agricultural Resources Board

(part of the Ministry of Agriculture) has overall

responsibility for fire management. lnformation

is typically required at two levels: (a) on an opera-

tional basis (mostly active fires and vegetation

status) to react to actual conditions and act

appropriately, and (b) to document the situation.

The latter is used to improve knowledge of the

potential problem, to assist strategic decision-

making, and eventually to assist in evaluating the

effectiveness of actions. Good archives are essen-

tial to monitor trends and evolution of fires and

burned areas. While there are mechanisms for


fire reporting in Botswana, the documentation of 8.18

fire events is incomplete. For example, when

vehicles are available, the extent of burned areas

is estimated by driving around them. This operation

is time consuming and can be highly inaccurate

due to the heterogeneity of the burned areas.

Countries as large as Botswana, that contain

extensive areas of fire-prone rangeland, could not

justify the expenditure needed to install and main-

tain observational networks for collecting fire

information regularly at a national level (over the

entire territory). This is where remote sensing

data such as NOAA-AVHRR can be used as a sub-

stitute to fill in gaps where data is not available,

or at least to prioritise action where resources

are limited.

While the Government of Botswana intends

to improve strategies on fire management nation-

ally, current resources are limited. When unde-

sired active fires are discovered in the field in

time, village resources are combined with any

available district resources to combat the fire.

However, most effort is put into prevention, such

as the establishment and maintenance of fire-

breaks. There are currently 6000 km of firebreaks

in Botswana, which require regular maintenance

to be effective, and there are plans to bring the

extent of the network up to I 0 000 km.

Figure 8.1 6. AVHRR-based burned area map (June-October 1998). Each colour corresponds t o the date at which the area

burned.

Figure 8.17. AVHRR-based fire frequency map for Botswana 1996-1998 (Ntabeni, 1999). Figure 8.1 8. AVHRR-based prototype fire danger map for Botswana (Ntabeni, 1999).


8.6.3.3 NOAA-AVHRR Products

and their uses

The Government of Botswana, through the

Department of Meteorological Services (DMS),

requested the UK Department for International

~evelopment's assistance in developing the

capacity to access and utilise data from satellite

remote sensing in order to complement the

monitoring of weather, climate, vegetation and

fires. With a PC-based NOAA receiver maintained

and run by local staff, DMS has been monitoring

vegetation status and burned areas since 1996,

working closely with the Botswana Rangeland

inventory Monitoring Project. They then feed the

information to decision makers in the Agricultural

Resources Board and Inter Ministerial Drought

Committee. While the data is potentially very

useful, it takes time and iterations of the products

that are delivered for product users to appreciate

the potential and product developers to meet

decision makers' requirements.

Vegetation status maps, as designed by Sannier

et al. ( 1998), have been used operationally by the

drought committee for the early identification of

problem areas. The usefulness of fire information

from NOAA-AVH RR was not directly perceived.

Initial burned area maps attracted attention of

decision-makers. Obtained by visual analysis and

digitisation on the screen, their production was

time consuming and the results missed most small

burned areas. Demand arose for an operational

approach to automatically detect burned areas

from the AVHRR data acquired daily. Current

knowledge in burned area detection was applied

to create an operational prototype to produce maps

automatically (Flasse, 1999). While providing very

sensible results (Figure 8.16), additional issues

rose, underlying the complexity of automated

burned area detection from AVHRR data.

As complete daily coverage of Botswana is

available, burned area products can be produced,

using the region and frequency that best suit

users' requirements. For example, daily data are

sometimes used to monitor evolution of large

fires lasting several days, while monthly syntheses

are used by the Central Statistics Department

of the Ministry of Finance. Perhaps of greater

importance is the accumulation of data on burned

areas over a number of years, particularly for

monitoring the frequency of fire occurrence

(Figure 8.17). This is of value to rangeland manage-

ment because of the strong link between the fre-

quency and impact of fire. It is also an important

indicator to contribute to risk assessment of fire

occurrence. In addition, firebreaks will be main-

tained with priority in areas where fuel load is

not reduced by fires in the past years.

While the burned area product from AVHRR

data, even thought not perfect, is slowly entering

into the routine of fire management, the Agricul-

tural Resources Board is increasingly interested

in the AVHRR data potential. For the 2000 fire

season, they requested and received information

from DMS on detected active fires, in near-real-

time.

Finally, local staff is currently working on a

first prototype method to produce ten-day fire

potential maps for Botswana (Ntabeni, 1999), an

example of which is given in Figure 8.18. The

model uses various inputs, such as AVHRR NDVI,

from the wet season as an indicator of fuel load,

AVHRR RGI (Relative Greenness Index) during


8.19 Monthly burned area for Senegal from 1993 to 1998

600 WG . . -

Nov Dec Jan Feb Mar Apr May

the dry season as an indicator of vegetation

status, AVHRR burned area maps, roads and settle-

ments, land use and vegetation maps.

The fire products obtained from NOAA-

AVHRR provide a very valuable contribution to

improved knowledge and fire management

decision-making. Some of the existing products

are already used operationally. The more decision

makers see and use these products, the more

they learn about their potential and gain useful

insights as to how it would best support them.

8.6.4 Senegal

8.6.4.1 The Fire Issues in Senegal

The fire season extends from October to May.

Overall, the critical fire period in Senegal is

variable between seasons. It depends on several

factors, including the rainfall (quantity and length),

the fuel production, vegetation status, spatial

distribution, awareness of population and pre-

scribed burning practices (CSE, 1999). Particularly

in the Sahelian part of Senegal, fire occurrence

contributes to increasing pressure on agricultural

and rangeland systems through the destruction of

natural pasture equilibrium and the weakening of

r- '?

. .

agricultural land. The fire activity on the forests of

the southern part of the country results in a de-

crease in the wood productivity as well as a threat

on regeneration. Bush fire is one of the main

causes for the degradation of natural resources,

and often results in changes in vegetation as well

as the living conditions of the local populations.

These fires are generally characterised by

their frequency, their unpredictability, and their

variable intensity. These are linked in turn to the

state of the vegetation or fuel, the variety of

existing ways to exploit natural resources and to

the social conditions of local communities (Mbow,

1997; CSE, 1999).

The Government of Senegal put means in place

to fight the bush fires in the principal eco-

geographic zones of the country. Where organ-

ised, communities are provided with equipment

to fight the fires (active activities). Passive activi-

ties consist more of awareness campaigns to help

local communities avoid conditions favourable to

wild fires. In the Sahelian part of the country a

network of firebreaks was established in order to

reduce fire spread or even stop it in areas where

Figure 8.1 9. Monthly burned area for Senegal between 1993 and 1998 (source: CSE). Figure 8.20. Fire frequency in Senegal between 1996 and 1998 (source: CSE).


nazciral obstacles are rare. However, firebreaks

tend not to be maintained because of the cost

associated with the operations and their efficiency

is therfore reduced. One of the most important

stmtegies was introduced in 1965, consisting of

eai-ly season prescribed fires to reduce fuel load

arid therefore to prevent late fires (much larger,

more difficult to control and more destructive).

When applied appropriately, this method is very

efficient. The effectiveness of those strategies goes

togaher with an appropriate use of fire informa-

tion by the public services as well as the general

public. Traditionally, fire information consisted of

field reports of observed or fought fires. How-

ever, the delay in providing active fire information

is usually proportional to the distance between

the fire and the fire-fighting unit. In addition, the

spatial and temporal variability of the bush fire

activity often exceeds the current means to react.

8.6.4.2 Fire Monitoring

The estimation of burned areas and the fire fre-

quency are fundamental aspects to try to manage

the natural resources with respect to fire activity.

To complement traditional fire information and

government initiatives, the Centre de Suivi

~ c o l o ~ i ~ u e (CSE) of Dakar has implemented a

methodology to monitor fire activity using NOAA-

AVHRR satellite data received locally through their

own station installed in 1992. Fires are identified

using night-time imagery and simple threshold

techniques. In 1999 the CSE became one of the

nodes of the World Fire Web of the Joint Research

Centre (EU), which provides operationally active

fire information during daytime. The fire infor-

mation is introduced, via GIs, into its geographical

context, allowing improved interpretation and

therefore support to management.

While not exhaustive, it is now well accepted

that fire information obtained from AVHRR data

provides good indication of the fire activity over

large territory. Initially only the territory of

Senegal was covered, and information operationally

used by the Forestry, Waters, Hunting and Con-

servation Department, the Livestock Directorate

and the National Parks Directorate. Now CSE

monitoring activities are also contributing to

providing fire information for the neighbouring

countries.

8.6.4.3 Fire Activity During 1 993- 1 998

The analysis of the fire information over the

period 1993 to 1998 resulted in a description of

the fire activity in Senegal and the identification

of important issues.

The problematic period is often January to

February, when remaining high volume of vegeta-

tion is senescent and the weather very dry. During

that period large uncontrolled wild fires can be very

destructive. The spatial and temporal distribution

is generally heterogeneous and variable.

Figure 8.19 illustrates this variability and clearly

indicates peak fire activity in February 1994 and

January 1996. In 1994, these can probably be ex-

plained by the absence of prescribed fires. In 1 99516,

high rainfall continued until November, increasing

fuel loads and delaying the application of prescribed

fires. This is believed to be the main cause of the

recrudescent fires of January 1996.

The regression period is characterised by a sharp

decrease in area burned due to a corresponding

decrease in fuel load (vegetation already burned


or grazed). It takes place between March and May.

However, during that period, the data quality is

affected by increased cloud cover over the south

of the country. It is also in this area that the late

season fires are observed. Slash and burn agri-

culture is practised in the south-east, which

usually sees an increase in fire activity at the end

of the dry season, corresponding to field prepa-

ration for the coming crop season.

Spatially, as is illustrated in Figure 8.20, fire

activity occurs in the centre, south and south-west.

Most of the fire activity takes place in the regions of

Kolda, Tamba and Ziguinchor, because of their con-

tinuous herbaceous cover combined with human

activities such as honey and gum collection, hunting

and charcoal production. In the north of the country,

the lower biomass is usually used by the cattle and

fire activity is consequently very low.

8.6.4.4 Partners

The Forestry, Waters, Hunting and Conservation

Department (DEFCCS) is responsible for fire

control activities. It puts into place the policy of

fire prevention and fire fighting, but it lacks means.

CSE provides the DEFCCS with fire information

that is used to identify the locations and size of

fires. This information helps managers t o focus

on fragile areas as well as to allocate resources

appropriately.

The Livestock Directorate is responsible for

the management of rangeland, indispensable to

feed the cattle. Since bush fires destroy valuable

forage, often resulting in over-grazing, fire infor-

mation provided by the CSE is used to better

organise pastoral movements. However, with the

recent decentralisation processes, fire control

responsibilities and pastoral movements have

been transferred to local communities. Unfortu-

nately, the newness of the process combined with

scare resources prevents local communities from

playing their role suitably.

The National Parks Directorate receives from

CSE, in near-real time, information on the fire

activity in the Parc National de Niokolo Koba,

important heritage for its flora and fauna. Fire

information is used by the managers of the park to

assess awareness campaigns in the neighbowing

villages, as well as to prescribe burning activities.

Finally, through international collaborations,

the impact of radio campaigns in Guinea has been

assessed using remote sensing fire information

from CSE. The study showed a reduction of fire

activity in areas covered by the radio campaigns

(Kane, 1997). Clearly the fire issues are inter-

disciplinary and must be tackled as a common and

integrated effort where satellite data can positively

contribute to efficient fire management.

8.6.5 Ethiopia

In February to March 2000, Ethiopia experienced

damaging large forest fire events impacting on

the only remaining significant natural forest

areas of the Eastern Highlands. These areas are an

important part of the Protected Areas System of

Ethiopia, for their biodiversity as well as resources

for the local communities. In Bale alone, these

fires affected 45 forest priority areas, damaged

53 000 ha of forest and 1000 ha of wild coffee,

killed 30 head of livestock and 49 of wildlife,

destroyed over 5000 beehives and 43 houses.

Whether for the short term or the longer term,

these fire events clearly affected resources


imi:jirant to people's livelihoods. However, the

wirjei- issues linked to fire are complex, and people

are tisually at their origin. Mainly pastoralists,

farn-,ers, hunters and honey gatherers are starting

the fires in Ethiopia. There i s an increased

den--and for farmland to sustain the livelihood of

t h ~ fast growing population. Given the backward

agricultural techniques with low productivity,

expaasion of farmland is the only option for many

famiiies. In addition, activities by immigrants with

no cult-ural affiliation with the forests, and there-

fore little knowledge on the ecology of forest,

represent an important threat to the sustainability

of tl-ieir main natural resources. Through the Global

Fire Monitoring Center, the Government of Ethio-

pia received emergency support from the inter-

national community (Germany, USA, South Africa,

Canada, UNEP). In particular, the United States

National Oceanic and Atmospheric Administration

(NOM), National Environmental Satellite, Data,

and hiformation Service (NESDIS), International

and interagency Affairs Office, on the request of

the Government of Ethiopia through i ts embassy

in Addis Ababa, provided the following remote

sensing fire information:

Images from the DMSP (US Air Force Defence

Meteorological Satellite Program) images at

2.7 km resolution for the East Africa region.

A special survey area where the fires

occurred (Goba and Shakiso Regions: 5-9"N,

38-42"E), were produced daily (weekdays).

Images from the NOAA-AVHRR (Advanced

Very High Resolution Radiometer) at I km

resolution (recorded onto NOAA- I4 space-

craft), from 8 to I 0 March 2000 and occasion-

ally later. Restrictions were due to the fact

that the satellite's orbital track changed and

the spacecraft did not image directly over

Ethiopia due to other commitments for

recording I x I km resolution data.

Fire maps were particularly useful during the

emergency period to assess the evolution of the

situation and help flying crews identifying active

fire locations.

8.6.5.1 Fire Information and Management

The Government of Ethiopia recently decided to start

the development of integrated fire management

strategies in order to prepare for catastrophes,

and most importantly to prevent them through

improved awareness and integrated fire manage-

ment in order to benefit in the long term both the

local communities and the forest ecosystem. The

first step was a Round Table Conference on forest

fire management in September 2000, in order to

learn from the past events and from experiences in

other countries and so define recommendations

for the coming years (Ministry of Agriculture,

Ethiopia, with GTZ and GFMC 200 1).

The Round Table clearly recognised the im-

portance of taking into account all aspects relat-

ing to the fires, with particular attention given to

the status of the people initiating them. The

development of a fire information system and

the use of fire remote sensing capabilities were

recommended, and particular emphasis was

placed on the existing remote sensing capabili-

ties of the Ethiopian National Meteorological Sew-

ices Agency (NMSA), in Addis Ababa (Flasse,

2000).


8.6.5.2 NOAA-AVHRR Products

and their uses

The NMSA hosts receiving stations for the NOAA-

AVHRR and Meteosat satellites. Initially imple-

mented in 1990 in support of drought prepared-

ness (Tsegaye et al., I995), the same data can be

interpreted for fire management. NMSA already

operates the systems and collect satellite data

daily. In December 2000 - further to the Round

Table recommendations - those capabilities were

upgraded to cover fire information. Both NMSA

and Ministry of Agriculture staff were trained to

use active fire detection software, to extract

active fire locations and to integrate the informa-

tion into the forestry GIs.

The forest and fire community in Ethiopia is

now starting to build on this new expertise to

benefit from timely and national fire information

from satellite data. An example is given in Figure

8.2 1 .

8.7 FUTURE EXPECTATIONS

The NOAA-AVHRR, Landsat and SPOT satellite

systems have been the workhorses for land cover

applications until recently. Several new remote

sensors have been developed for the "new

generation" of satellites reflecting the trend in

remote sensing towards increasingly specific

applications and higher sensor resolution. This

leads to a tremendous increase in the amount of

data in need of processing and storage, but con-

current advances in computer hardware and soft-

ware are keeping pace with requirements. There

is greater emphasis on making remote sensing

products more accessible to a wider range of

users. This means that, not only will raw imagery

be available, but also derived information products

that are more user-friendly, for those without a

remote sensing background. Some of these new

data products include maps of net primary pro-

duction, leaf area index, land cover change, and

fire. Furthermore, new data and data products

are increasingly available for free (e.g. EOS data)

or at substantially lower cost than ever before (in

the case of Landsat 7).

The first and second Along Track Scanning

Radiometer (ATSR) instruments, ATSR- I and

ATSR-2, have been operating since 199 1 and 1995

on board the ERS- I and ERS-2 satellites, respec-

tively. The Advanced ATSR {AATSR) instrument

will be launched on ESA's Envisat platform in the

near future. ATSR-2 and AATSR have green, red

and NIR channels for vegetation monitoring, in

addition to the two SWlR and two TIR channels

on ATSR-I. Swath width is 500 km and spatial

resolution at nadir is I km. The key feature of

ATSR is that it can deliver both nadir and "along

Figure 8.21. Vegetation (NDVI) and active fire information from NOAA-AVHRR data over west Ethiopia.


traci:" views of the same surface location where

the fziter view passes through a longer atmos-

pheric path, thus enabling improved corrections

for a;-mospheric effects.

The Meteosat Second Generation (MSG) pro-

gramme will continue where the Meteosat pro-

gramme began in 1977, and will be particularly use-

ful for regional/continental-scale monitoring of fires,

like the AVHRR. In addition, it will provide

data every 15 minutes, allowing the monitoring

of fire progression and fire temporal distribution.

The MSG satellites will operate from geo-

stationary orbits, and provide multi-spectral

imagery in 12 spectral channels, at I km spatial

resolution in the visible channel and 3 km for the

others, 8 of which will be in the TIR. Most Na-

tional Meteorological Services in Africa are ex-

pected to be equipped with the relevant receivers,

allowing near-real time monitoring in-country.

The two principal EOS (Earth Observing

System - NASA) platforms are Terra (EOS AM-

I) and Aqua (EOS PM-I). Both Terra and Aqua

feature the Moderate Resolution lmaging

Spectrometer (MODIS). The MODlS instrument

on board Terra is considered more useful for land

surface applications due to i t s morning flyover

time, especially in the tropics, where clouds usu-

ally develop by afternoon. MODlS has a swath

width of 2330 km and a repeat cycle of one to two

days, which makes it the principal sensor for

monitoring the Earth system, replacing AVHRR,

but with some important improvements. The red

and NIR bands have a spatial resolution of 250 m,

allowing global NDVl information at much finer

resolution than with AVHRR. Bands 3-7 (500 m)

and 8-36 ( 1 000 m) provide additional data in the

short-wave infrared (SWIR) and thermal infrared

(TIR) wavelengths. MODIS data is contributing

substantially to global fire monitoring, along with

fire effects on land and atmospheric processes.

Furthermore, a range of MODlS active fire prod-

ucts are already freely available online (Annexure

I), and a 500m burned area product is being re-

fined and evaluated ready for general release.

A new type of sensor on board Terra i s

ASTER, which consists of three separate sub-

systems corresponding to three spectral regions:

Visible and Near lnfrared (VNIR), Shortwave

lnfrared (SWIR) and Thermal lnfrared (TIR). The

VNIR subsystem has three spectral bands in the

visible and NIR wavelengths, with 15 m spatial

resolution. The nadir-looking detector is com-

plemented by a backward-looking detector to

permit stereo viewing in the NIR band. The SWIR

subsystem features six spectral bands in the near-

IR region, with 30 m resolution. The TIR sub-

system has five bands in the thermal infrared

region, with 90 m resolution. ASTER'S 60 km

swath width gives it some continuity with SPOT,

and ASTER images are already proving useful for

detecting burned areas and capturing real-time

fires.

Another new sensor on Terra, the Multi-angle

lmaging SpectroRadiometer (MISR), features nine

widely spaced view angles for monitoring the

Earth's surface. This capability allows for the

improved extraction of quantitative parameters

describing the surface of the Earth through, for

example, the inversion of bi-directional reflect-

ance models. MISR provides coverage of the en-

tire Earth's surface in swaths 360 km wide by

20 000 km long, every nine days. Pixel size is

wda'iand Fire Management Handbook for Sub-Sahara Africa

250 m at nadir, and 275 m from the off-nadir

cameras.

An important consideration for monitoring land

cover change over dekadal (ten-day) time scales

is data continuity between satellites and satellite

systems. The Landsat satellite series began in

1972, and thus represents the longest available

time series. The most recent addition to the

Landsat programme was Landsat 7, which began

providing data in July 1999. Landsat 7 has an

Enhanced Thematic Mapper (ETM +) instrument

that features enhanced radiometric resolution

in the six TM channels, plus improved spatial

resolution in the thermal channel (60 m), plus a

panchromatic band with 15 m spatial resolution.

Similarly, SPOT imagery has been available since

1986, and the SPOT series was enhanced with

the launch of SPOT 5 in 2002. AVHRR also pro-

vides nearly 20 years' daily data over the whole

globe. Since 1998, the coarse-resolution SPOT-

VEGETATION sensor has provided another tool

comparable to AVHRRfor daily monitoring of global

vegetation at I I<m spatial resolution, in four spec-

tral bands (blue, red, NIR and SWIR).

In conclusion, the future is now with regard to

remote sensing of fires and fire effects on land

cover and landscape processes. Recent, dramatic

improvements in remote sensing and data pro-

cessing capabilities, data product availability and

internet access should lead to equally dramatic

improvements in remote detection, measurement

and monitoring of fires and fire effects.

8.8 ANNEXURE 1

INTRODUCTORY TEXTBOOKS

Wilkie, D.S. and1.T. Finn. 1996. Remote Sensing Imagery for

Natural Resources Monitoring - A Guide for First-Time Users.

Columbia University Press, New York.

Sabins, F.E, jr. 1996. Remote Sensing: Principles and Inter-

pretation. 3rd ed. Freeman, New York.

Lillesand, TM. and R.W. Kiefer. 1999. Remote Sensing and

Image Interpretat ion. 4th ed. John Wiley, New York.

Barrett, E.C. and L.F. Curtis. 1999. lntroduction to Environ-

mental Remote Sensing. 4th ed. Nelson Thornes, London.

Campbell, J.B. 1996. lntroduction to Remote Sensing. 2nd ed. Taylor & Francis, London.

Gibson, F? and C. Power. 2000. Introductory Remote Sensing

Principles and Concepts. Routledge, London.

STATE-OF-THE-ART REVIEW

Ahern, F., J.G. Goldammer and C. justice (eds.). 2001. Global and regional vegetation fire monitoring from space:

Planning a coordinated international effort. SPB Academic Publishing bv, The Hague.

Innovative Concepts and Methods in Fire Danger Estimation

(Proceedings of the 4th International Workshop on Remote Sensing and GIs Applications to Forest Fire Management, Ghent, Belgium, 5-7 June 2003). http://www.geogra.uah.es/

earsel/report I .html

USEFUL WEB PAGES O N REMOTE SENSING

The following are good places to start from as they contain lots of links to other remote sensing pages:

The Remote Sensing and Photogrammetry Society: http://www.rspsoc.org/

WWW Virtual Library: Remote Sensing http://www.vtt.fi/tte/research/tte I /tte 14/virtual/

Aqira

ARVI

ASAR

ASTER

EDC

Envisat

EOS


Universiteit Utrecht

http://\~w.frw.ruu.nl/nicegeo.html#~is

,r.,CRONYMS ABBREVIATIONS 4fdD EXPLANATIONS

Nrsf i

AM l

Aqua

A RV 1

ASAR

ASTER

ATSR

AVH RR

DPhC

DMSP

EDC

Envisat

EOS

EOSDiS

Advanced Along-Track Scanning Radiometer

(visiblelinfrared sensor on Envisat series,

successor of ATSR)

Active Microwave Instrument (SAR sensor on

ERS series)

EOS satellite (formerly known as EOS PM- I )

Atmospherically Resistant Vegetation lndex

Advanced Synthetic Aperture Radar (micro-

wave sensor on Envisat; successor of AMI)

Advanced Spaceborne Thermal Emission & Reflectance Radiometer (sensor on Terra

satellite)

website: http://asterweb.jpl.nasa.gov/

Along-Track Scanning Radiometer (visible/

infrared sensor on ERS series)

website: http://earthnet.esrin.esa.it

Advanced Very High Resolution Radiometer

(sensor on NOAA satellite series)

Distributed Active Archive Center (US data

collection points, these can usually be easily

accessed via the Internet)

Defense Meteorological Satellite Program

(USA)

EROS Data Center (part of United States

Geological Survey), data sales (hosts a DAAC)

Satellite, successor to ERS programme

Earth Observing System (NASA's Earth

Science satellite programme)

Earth Observing System Data and Information

System. The EOS Data Gateway provides a

central search and order tool for accessing a

wide variety of global Earth science data and

information held at 8 different EOSDIS data

centres and a growing number of international

data providers.

European Remote Sensing Satellite series

ETM+ Enhanced Thematic Mapper (sensor on

~arf8sat-7)

website: http://landsat7.usgs.gov/

Geostationary A type of satellite orbit (at 36 000 km

GEM1

GES

GFMC

GOES

H RV

H RVI R

lkonos

I RS

JERS

Landsat

LARST

LMlR

Meteosat

MIR

MlSR

MODIS

above the equator) where the motion of the

satellite matches the speed and direction of

the Earth's rotation so that the satellite

remains over a fixed point on the Earth's

surface. Also called geosynchronous.

Global Environment Monitoring lndex

GSFC (Goddard Space Flight Center) Earth

Sciences (hosts a DAAC with MODlS data)

Global Fire Monitoring Center

website: http://www.fire.uni-freiburg.de/

Geostationary Operational Environmental

Satellite - meteorological satellite programme

(USA)

High Resolution Visible (sensor on SPOT- I, -2 and -3)

High Resolution Visible Infrared (sensor on

SPOT-4 and -5)

Space lmaging EOSAT high resolution visible

satellite series

Indian Remote Sensing Satellite series

Japanese Earth Resources Satellite (visible/

near infrared and microwave sensors)

Land use studies satellite series (USA)

(variously carries sensors MSS, T M and

ETM +)

Local Applications of Remote Sensing

Techniques (former programme of Natural

Resources Institute)

Long Mid-Infrared (waveband or sensor

channel)

European meteorological satellite series

Mid-Infrared (waveband or sensor channel)

Multi-angle lmaging Spectro-Radiometer

(sensor on Terra satellite)

website: http://www-misr.jpl.nasa.gov/

Moderate Resolution lmaging Spectrometer

(sensor on Terra satellite) website (info):

http://modis.gsfc.nasa.gov/ and http://

edcdaac.usgs.gov/main. html


MSG

MSS

NASA

N DVl

N I R

NOAA

OrbView

PAN

Meteosat Second Generation (satellite series,

successor to Meteosat)

website: http://www.esa.int//msg/

Multispectral Scanner System (sensor on

Landsat series)

National Aeronautics and Space Administration

(USA)

Normalized Difference Vegetation lndex

Near lnfrared (waveband or sensor channel)

National Oceanic and Atmospheric Adminis-

tration (USA, also the name of their satellite

series)

Orblmage high resolution visible satellite

series

Panchromatic (often used to refer to sensor

data with a single waveband visible channel,

e.g. SPOT PAN)

Polar orbit An orbit where the satellite flies around the

Earth travelling approximately north to south

(or south to north) so that its path goes over

the polar regions.

Radarsat Canadian radar satellite (with SAR sensor)

SAC Satellite Applications Centre (South Africa,

data sales)

SAFNet

SAR

SAVl

SMlR

SPOT

SSMII

Terra

TIR

TM

TSAVl

USGS

VGT

VI S

XS

Southern African Fire Network

website: www.safnet.net

Synthetic Aperture Radar (microwave sensor)

Soil Adjusted Vegetation lndex

Short Mid-Infrared (waveband or sensor

channel)

Satellite Pour I'Observation de la Terre

(SPOTlmage (French) satellite series)

website: http://www.spot.com

Special Sensor Microwave Imager (passive

microwave sensor on DMSP)

EOS satellite (formerly known as EOS AM- I)

Thermal lnfrared (waveband or sensor

channel)

Thematic Mapper (sensor on Landsat-4 and -5)

Transformed Soil Adjusted Vegetation lndex

United States Geological Survey (includes

EROS Data Center, data sales)

Vegetation (AVHRR-like sensor on SPOT-4

and -5)

Visible (waveband or sensor channel)

Often used t o refer t o SPOT multispectral

data


Apado, I., E. Chuvieco and J. Salas. 200 1 . Assessment of forest fire danger conditions in Andalucia from remote sr-nsing images (NOAA) and meteorological indexes. In: Proceedings of the 3rd lnternational Workshop on Remote

Sensing and GIS applications to Forest Fire Management:

h!ew methods and sensors ( E . Chuvieco, and I? Martin, eds), 45-48. European Association of Remote sensing Laboratories (EARSeL), Paris.

Ahern, E, J.M. Gregoire and C. Justice. 2000. Forest fire monitoring and Mapping: Acomponent of Global Observa- tion of Forest Cover. Report of a Workshop, 3-5 Novem- ber1999, JRC, Ispra, Italy, EUR 19588 EN, 25613.

Belwai-d, A.S., J. M. Gregoire, G. D'Souza, S. Trigg, M. Hawkes, j. VI. Brustet, D. Serga, J. L. Tireford, J. M. Charlot and R. Vuattoux. 1993. In-situ, real-time fire detection using Ik!OAA/AVHRR data In: Proceedings of the 6th European AWRR Data Users' Meeting, 29 June to 2 July 1993, Belgirate, Italy. Eumetsat, Darmstadt, Germany, 333-339.

Bond, WJ. and B.W. van Wilgen. 1996. Fire and Plants.

Population and Community Biology Series 14. Chapman and Hall, London, 263 p.

Brockett, B.H., H.C. Biggs and B.W. van Wilgen. 200 1 . A patch- mosaic burning system for conservation areas in southern African savanna. Int. j. Wildland Fire 10, 1 69- 1 83.

Burgan, R.E. 1996. Use of remotely-sensed data for fire danger estimation. EARSeL Advances in Remote Sensing,

4(4), 18.

Burgan RE, R.W. Klaver and J.M. Klaver. 1998. Fuel models and fire potential from satellite and surface observations. International journal of Wildlond Fire 8,159- 1 70.

Campbell, J.B. 1987. Introduction to remote sensing. Guildford Press, New York.

Carlson, J. D. 200 1. Review of user needs in operational fire danger estimation: The Oklahoma example. In: Proceedings

o j the 3rd International Workshop on Remote Sensing and

GIS applications to Forest Fire Management: New methods

and sensors (E. Chuvieco, and I? Martin, eds.), European

Association of Remote sensing Laboratories (EARSeL).

Ceccato, R, S. Flasse, S. Tarantola, S. Jacquemoud and J. M.

GrBgoire. 200 I (a). Detecting vegetation leaf water content using reflectance in the optical domain. Remote

Sensing of Environment 77( I), 22-3 3.

Ceccato, t?, N. Gobron, S. Flasse, B. Pinty and S. Tarantola.

2002(b). Designing a spectral index to estimate vegetation water content from remote sensing data. Part I:

Theoretical approach. Remote Sensing of Environment, 82 (2-3), 188- 197.

Ceccato, F, S. Flasse and J. M. Gregoire. 2002(a). Designing a spectral index to estimate vegetation water content from remote sensing data. Part II: Validation and applications. Remote Sensing of Environment, 82 (2-3), 198-207.

Cheney, I? and A. Sullivan. 1997. Grassfires fuel, weather and

fire behaviour. CSlRO Publishing, 102 p.

Christensen, N.L. 1 98 1. Fire regimes in southeastern ecosystems. In: Proceedings of the Conference: Fire Regimes

and Ecosystem Properties, December 1 1- 15, 1978. U.S. Department of Agriculture General Technical Report WO- 26, 1 12- 136.

Congalton, R. and K. Green. 1999. Assessing the accuracy of

remotely sensed data: Principles and practices. Florida, CRC Press, Florida.

Cracknell, A.I? 1997. The Advanced Very High Resolution Radi-

ometer (AVHRR). Taylor and Francis, London.

CSE - Centre de Suivi Ecologique. 1999. Etude diagnostic sur les feux de brousse dans les regions de Kolda et Tarnbacounda. Strategies alternatives. Rapport de consultation pour le compte du Programme de Gestion Durable et Participative des Energies Traditionnelles et de Substitution, 63p.

Downey, I.D. 1994. Remote sensing activities at the Natural Resources Institute for measurement of geophysical parameters. In: The Determination of Geophysical Parameters

from Space, N.E. Fancey, I.D. Gardiner and R.A. Vaughan

(eds.), SUSSP Publications, 33-44.

Downey, I.D. 2000. Data feast, information famine - Opera- tional remote sensing in developing countries.

Geolnformatics, July/August 2000, 24-26.

Du Plessis, W.R 1997. Refinements to the burning strategy in

the Etosha National Park, Namibia. Koedoe 40(1), 63-76.

Du, Y, ET Ulaby and M.C. Dobson. 2000. Sensitivity to Soil Moisture By Active and Passive Microwave Sensors. IEEE

Transactions on Geoscience and Remote Sensing 38(1), 105-1 14.

Edwards, A., G. Allan, C. Yates, C. Hempel and R Ryan. 1999. A comparative assessment of fire mapping techniques and user interpretations using Landsat imagery. Proceedings of the Australian Bushfire Conference, Albury, July 1999.


Eva, H. and E.F. Lambin. 1998. Burnt area mapping in Central Africa using ATSR data. Int. 1. Remote Sensing 19, 3473- 3497.

~verson. C.S. and M. Thompson. 1993. Development of spectral-biomass models for mapping and monitoring montane grassland resources. CSlR internal Report FOR-I 345 p.

Flasse, S. 1 999. Semi-Automatic AVHRR Burn Scar Detection for

Botswana Department of Meteorological Services, Prototype

Version 2.0, User Guide. University of Greenwich, 24 p.

Flasse, S. 2000. Remote sensing of vegetation fires and its contribution to a national fire information system. In: Ethiopia Round Table Workshop on Forest Fire Manage- ment (1 9-20 September 2000) (Ministry of Agriculture, Ethiopia, with GTZ and GFMC, ed.), Proceedings, 85-94. Ministry of Agriculture, Addis Ababa, 166 p.

Flasse, S. and M.M., Verstraete. 1994. Monitoring the environment with vegetation indices: comparison of NDVl and GEM1 using AVHRR data over Africa. In: Vegetation,

Modelling and Climatic Change Efects (E Veroustreate, and R. Ceulemans ed.), 107- 135. SPB Academic Publish- ing bv. The Hague.

Flasse, S.l? and F? Ceccato. 1996. A contextual algorithm for AVHRR fire detection. Int. J. Remote Sensing 1 7(2), 4 19- 424.

Flasse, S.P, I? Ceccato, I.D. Downey, J.B. Williams, F! Navarro and M.A. Raimadoya. 1998. Remote sensing and GIs tools to support vegetation fire management in developing countries. In: Proceedings of the 1 3th Conference on Fire and Forest Meteorology, 27-3 1 October 1996, Lorne, Australia, lnternational Association of Wildland Fire, USA, 209-2 14.

Gibson, Pj. and C. H. Power. 2000. Introductory remote sensing:

digital image processing and applications. London, Routledge.

Goldammer, J.G. 1998. Development of national fire policy and guidelines on fire management in Namibia. Report, Directorate of Forestry, Ministry of Environment and Tourism, Republic of Namibia.

Goldammer, J.G. 1999. Namibia Round Table Meeting on Fire, Windhoek 10- 1 1 November 1998. Report to Di- rectorate of Forestry, Ministry of Environment and Tour- ism, Republic of Namibia.

Grimes, D.I.F., R. Bonifacio and H.R.L. Loftie. 1998. Rainfall

Estimation Workbook. Chatham, UK. Natural Resources Institute.

Herman, A., V. Kumar, I? Arkin and J. Kousky (no p~bl icaio~

date). Objectively Determined 10 Day African Rainfall Estimates Created For Famine Early Warning Systems, published on the web at http:Nedcintl.cr.usgs.gov/adds/

datdrfedlrfedtext. html.

Hetherington, D.S. 1997. Geographic information and remote sensing tools to assist fire management protocols in the National Parks of South Africa. MSc thesis. Universiry

of Greenwich, United Kingdom.

Hetherington, D.S. 1998. The urilisation of satellite remote

sensing and GIs techniques in the generation of post fire alteration mapping in the Kruger National Park, South Africa. In: ASPRS-RTI 1998 Annual Conference Pro- ceedings, ASPRS, Bethesda, Maryland.

Hudak, A.T and C.A. Wessman. 1998. Textural analysis of historical aerial photography to characterize woody plant encroachment in South African savanna. Remote Sensing

of Environment 66(3), 3 17-330.

Hudak, A. T. and Brockett, B. H., in press, Mapping fire scars in a southern African savanna using Landsat imagery. h t .

1. of Remote Sensing.

Hudak, A.T. and C.A. Wessman. 2001. Textural analysis of high resolution imagery to quantify bush encroachment in Madikwe Game Reserve, South Africa, 1 955- 1996. Int.].

Remote Sensing 22,273 1-2740.

Hudak, A.T., B.H. Brockett and C.A. Wessman. 1998. Fire scar mapping in a southern African savanna. In: Proceed- ings of the lnternational Geoscience and Remote Sensing Symposium, Seattle, Washington, CD-ROM.

Jackson, R.D., S.B. Idso, R.J. Reginato and F! j. Pinter. 1981, Canopy temperature as a crop water stress indicator. Water Research 1 7, 1 1 33- 1 1 38.

Jacques de Dixmude, A., F? Navarro, S. Flasse, I. Downey, C. Sear, J. Williams, P Ceccato, R. Alvarez, E Uriarte, Z. ZQfiiga, A. Ramos and I. Humphrey. 1999. The use of

low spatial resolution remote sensing for fire monitorin in Nicaragua: A survey of three successive burnings sea

sons. lnternational Forest Fire News 20, 64-72. (http:ll www.ruf.uni-freiburg.de/fireglobe/other-rep/research/ni/

ni-re- l .htm)

Jakubauskas, M.E., K. F! Lulla and PW. Mausel. 1 990. Assess-

ment of vegetation change in a fire-altered forest la scape. Photogrammetric Engineering and Remote Se

56, 37 1-377.

(

Lefs k <


justice, C., J.P Malingeau and A.W. Setzer. 1993. Satellite remote sensing of fires: potential and limitations. In: Fire

!,-I the Environment: The ecological, atmospheric, and cli-

motic importance ofvegetation fires (PJ. Crutzen, and J.G.

Goldammer, eds.), 77-88. John Wiley, New York.

Kane, R 1997. Suivi satellitaire des feux en Guinee et de I'impact des campagnes de sensibilisation de la radio rurale. in: Comptes Rendus du Xieme Congres Forestier Mondial,

Antalya, Turquie 13-22 Octobre 1997.

Kaufman, YJ., A. Setzer, C. Justice, C.J. Tucker, M.C. Pereira 2nd I. Fung. 1990. Remote sensing of biomass burning in the tropics. In: Fire in tropical biota. Ecosystem processes and global challenges U.G. Goldammer, ed.), 37 1 - 399. Ecological Studies 84, Springer-Verlag, Berlin, 497 p.

Kaufman, Y.J., C.O. Justice, L.P Flynn, 1.0. Kendall, E.M. Prins, L. Giglio, D.E. Ward, W.P Menzel, A.W. Setzer. 1998. Potential global fire monitoring from EOS-MODIS.

J. Geophys. Res. D 103, 322 15-32238.

Kilgorz, B.M. 198 1. Fire in ecosystem distribution and structure: western forests and scrublands. In: Proceedings of the conference: Fire Regimes and Ecosystem Properties, December I I - 15, 1978. U.S. Department of Agriculture General Technical Report WO-26,58-89.

Koutsias, N., M. Karteris, A. Fern6ndez-Placios, C. Navarro, 1. Jurrado, R. Navarro and A. Lobo. 1999. Burnt land mapping at a local scale. In: Remote Sensing of Large Wildfires

in the European Mediterranean Basin (E. Chuvieco, ed.), 157- i 87. Springer-Verlag, Berlin.

Le Roux, J.L. 2000. Fire scar mapping in north-eastern Namibia. Report to the Directorate of Forestry, Ministry of Environment and Tourism, Namibia.

Le Roux, J.L. 200 1. Mapping fire scars and estimating burned area acreage for Namibia - 200010 I. Report to the Atlas of Namibia project, Windhoek, Namibia.

Lefslq, M.A., D. Harding, W.B. Cohen, G. Parker and H.H. Shugart. 1999(a). Surface Lidar Remote Sensing of Basal Area and Biomass in Deciduous Forests of Eastern Mary- land, USA. Remote Sensing of Environment 67,83-98.

Lefsky, M.A., W.B. Cohen, S.A Acker, G.G. Parker, TA.

Spies and D. Harding. 1999(b). Lidar remote sensing of the canopy structure and biophysical properties of

Douglas-fir western hemlock forests. Remote Sensing of Environment 70, 339-36 I.

Lillesand, M. and R.W. Kiefer. 2000. Remote sensing and image

interpretation. 4th ed. John Wiley, New York.

Martin Pilar, M., F! Ceccato, ST Flasse and I. Downey. 1999. Fire detection and fire growth monitoring using satellite data. In: Remote Sensing of Large Wildfires in the European

Mediterranean Basin (E. Chuvieco, ed.), I0 I - 1 22. Springer- Verlag, Berlin.

Mbow, C. 1 997. Methodologie d'identification et de Cartographie des feux de brousse par teledetection, utilisation des images satellitaires NOAA-AVHRR et IANDSAT Dakar UCAD - Memoire de fin d'etude pour I'obtention du DEA en sciences de I'environnement, 88 p.

Mendelsohn, J. and C. Roberts. 1997. An environmental profile

and atlas of Caprivi. Windhoek, Ministry of Environment and Tourism, Environmental Profiles Project, 44 p.

Mendelsohn, J., S. Obeid and C. Roberts. 2000. A Profile of

North-Central Namibia. Directorate of Environmental Affairs, Windhoek.

Milford, J.R., VD. McDougall and G. Dugdale. 1996. Rahfall estimation from cold cloud duration: experience of the TAMSAT group in West Africa. In: Validation problems of

rainfall estimation methods by satellite in intertropical

Africa (B. Guillot, ed.). ORSTOM. Publ. in Colloques et Seminaires, Proceedings of the Niamey Workshop, December 1994.

Ministry of Agriculture, Ethiopia, with GTZ and GFMC (ed.).

200 1 . Round Table Conference on Integrated Forest Fire Management in Ethiopia. Proceedings of the Work- shop, Addis Ababa, Ethiopia, 19-20 September 2000. Ministry of Agriculture, Addis Ababa, 166 p.

Miura, T, A.R. Huete, W.J.D. van Leeuwen and K. Didan. 1998. Vegetation detection through smoke-filled AVlRlS images: an assessment using MODIS band passes. I. Geophys. Res. 103, 3200 1-320 1 1 .

Moghaddam, M. and S.S. Saatchi. 1999. Monitoring Tree Moisture Using an Estimation Algorithm Applied to SAR Data from BOREAS. lEEE Transactions on Geoscience

and Remote Sensing 1 7,90 1-9 16.

Moran, M.S., TR. Clarke, Y lnoue and A. Vidal. 1994. Esti- mation crop water deficit using the relation between surface-air temperature and spectral vegetation index.

Remote Sensing of Environment 49,246-263.

Ntabeni, T 1 999. Fire potential modelling using NOAA-AVHRR data and other spatial data of Botswana. MSc Disserta- tion, University of Greenwich, 83 p.


Parr, C.L. and B.H. Brockett. 1999. Patch-mosaic burning: a

new paradigm for savanna fire management in protected areas? Koedoe 42(2), 1 1 7- 1 30.

Pereira,J.M. 1999. A comparative evaluation of NOAAJAVHRR vegetation indices for burned surface detection and mapping. IEEE Transactions on Geoscience and Remote Sensing

37, 2 17-226.

Pereira, J.M.C., A.C.L. Sa, A.M.O. Sousa, J.M.N. Silva, T.N. Santos and J.M.B. Carrieras. 1999(a). Spectral charac- terisation and discrimination of burnt areas. In: Remote

Sensing of Large Wildfires in the European Mediterranean

Basin (E. Chuvieco, ed.), 1 23- 1 38. Springer-Verlag, Berlin.

Pereira, J.M.C., A.M.O. Sousa, A.C.L. Sh, M. Pilar Martin, and E. Chuvieco. 1999(b). Regional scale burnt area mapping in Southern Europe using NOAA-AVHRR I km data. In: Remote Sensing of Large Wildfires in the European

Mediterranean Basin (E. Chuvieco, ed.), 139- 155. Springer- Verlag, Berlin.

Pereira, J.M.C., S. Flasse, A. Hoffman, J.A.R. Pereira, F. Gonzalez-Alonso and S. Trigg. 2000. Operational use of remote sensing for fire monitoring and management: regional case studies. In: Forest Fire Monitoring and Mapping: A Component of Global Observation of Forest Cover, Report of a Workshop, November 3-5 1999, Joint Research Centre, Italy, E Ahern, J-M. Gregoire and C. Justice, ed. European Commission -Joint Research Centre, EUR 19588 EN, 98- 1 10.

Pereira, M.C. and A.W. Setzer. 1993. Spectral characteristics of fire scars in Landsat 5 TM images of Amazonia. Int. j. Remote Sensing 14, 206 1-2078.

Rasmussen, M.S. 1998. Developing simple, operational, con- sistent NDVI-vegetation models by applying environmental and climatic information: Part I. Assessment of net primary production Int. j. Remote Sensing 1 9,97- 1 1 7.

Salvador, R., J. Valeriano, X. Pons and R. Diaz-Delgado. 2000. A semi-automatic methodology to detect fire scars in shrubs and evergreen forest with Landsat MSS time series. Int. 1. Remote Sensing 2 1 , 655-67 1 .

Sannier, C. 1999. Strategic monitoring of crop yields and rangeland conditions in Southern Africa with Remote Sens- ing. PhD Thesis, Cranfield University: Silsoe College.

Sannier, C.A.D., J.C. Taylor, W. du Plessis and K. Campbell.

1 998. Real-time vegetation monitoring with NOAA-AVHRR in Southern Africa for wildlife management and food secu-

rity assessment. Int. j. Remote Sensing 19, 62 1-639.

Sannier, C.A.D., J.C. Taylor and W. du Plessis. 2002. Real..

Time Monitoring of Vegetation Biomass with NOAA.. AVHRR in Etosha National Park, Namibia, for fire risk assessment. Int. j. Remote Sensing 23, 7 1-89.

Sarabandi, K. 1997. Delta k-radar equivalent of interfer~rnetri~

SARs: A theoretical study for determination of vegetation height. lEEE Transactions on Geoscience and Remote Sensing

35, 1267- 1276.

Schmugge, T. 1978. Remote sensing of surface soil moisture.

j. Appl. Meteorology 17, 1549- 1557.

Scholes, R.J. and B.H. Walker. 1993. Synthesis of the N$sv/eY

Study. Cambridge University Press, Cambridge, UK.

Stander, PE., T.B. Nott and M.T. Mentis. 1993. Proposed burning strategy for a semi-arid African savanna. AfrcanJ

Ecology 3 1 , 282-289.

Stronach, N.R.H. and S.J. McNaughton. 1989. Grassland fire dynamics in the Serengeti Ecosystem, and a potential method of retrospectively estimating fire enerwj. A#/.

Ecology 26, 1 025- 1033.

Thompson, M. 1990. Pilot study to evaluate multitemporal

LANDSAT Multispectral Scanner data for mapping seasonal burns in woody savanna type vegetation. SA For- estry Research Institute Centre Report P2/89.

Thompson, M. W. 1993. Quantitative biomass monitoringand fire severity mapping techniques in savanna environments using Landsat Thematic Mapper imagery. External Con- tract Report FOR-DEA 587, Division of Water, Environ- ment and Forest Technology, CSIR, Pretoria.

Thompson, M.W. and E.R. Vink. 1997. Fire scar mapping in Pilanesberg National Park. External Contract Report ENV/ PIC11 97 157, CSIR, Pretoria.

Thompson, M. and E.R. Vink. 200 1 . Biomass Determination for predictive Burn Risk Assessment. CSIR Externd Client Report ENVIPIC 200 1-0 1 3, March 200 1.

Trigg, S.N. 1996. Fire monitoring in the Caprivi. Report to Directorate of Environmental Affaris, Ministry of Envi- ronment and Tourism, Namibia.

Trigg, S. 1997. Fire scar mapping in Northern Namibia. Re-

port to Directorate of Forestry, Ministry of Environment and Tourism, Namibia.

Trigg, S. and S. Flasse. 2000. Characterising the spectral- temporal response of burned savannah using in sit spectroradiometry and infrared thermometry. Int. j. Re

mote Sensing 2 1, 3 16 1-3 1 68.

7

T

li

VE

Vi

Vii

Wi

Xie


Tr@. S. and S. Flasse. 200 1. An evaluation of different bi- spectral spaces for discriminating burned shrub-savannah. In:. j. Remote Sensing (in print).

Tr@, S. and J. le Roux. 2001. Hot Spot Contribution to the FA0 Special Report on Forest Fires. In press.

Trig, S . et al. (in preparation). Developing reference data for the accuracy assessment of burned area maps made from coarse spatial resolution, high temporal resolution data.

Trollope, W.S.W. 1992. Fire behaviour and its significance in

burning as aveld management practise. In: Prestige Farmers

Day Proceedings (C.R. Hurt and I?j.K. Zacharias, eds.). Grassland Society of Southern Africa, Scottsville.

Trollope, W.S.W. and L.A. Trollope. 1999. Technical review of the integrated forest fire management component of the Namibia-Finland Forestry Programme in the East Caprivi Region of Namibia. Report, Directorate of Forestry, Min- istr\/ of Environment and Tourism, Republic of Namibia.

TABLE OF SOME SATELLITE SENSORS AND DATA PROVIDERS

(see overleaf)

5egaye Tadesse, C.B. Sear, T Dinku and S. Flasse. 1995. The impact of direct reception of satellite data on a small African meteorological service: operational use of N O M AVHRR and METEOSAT products in Ethiopia. In: Pro- ceedings of the 1995 Meteorological Satellite Data Users Conference, Winchester, UK, 4-8 September 1995, EUMETSAT, Germany, 485-489.

Tucker, C.J. 1980. Remote Sensing of Leaf Water Content in the Near Infrared. Remote Sensing of Environment 10, 23-32.

Verstraete, M.M. and B. Pinty. 1996. Designing optimal spectral indexes for remote sensing applications. lEEE Transactions

on Geoscience and Remote Sensing 34, 1 254- 1 265.

Vidal, A., E Pinglo, H. Durand, C. Devaux-Ros and A. Maillet.

1994. Evaluation of a Temporal Fire Risk Index in Medi- terranean Forests from NOAAThermal IR. Remote Sensing

of Environment 49, 296-303.

Viedma, O., J.Melia, D. Segarra and j. Garcia-Haro. 1997. Modelling rates of ecosystem recovery after fires by using Landsat TM data. Remote Sensing of Environment 6 1, 383-398.

Wihns, J. 1 999. Towards coherence in developmental decision- making: the decision support roles of remote sensing and GIs - Lessons from the IARST approach. In: Decision tools

for sustainable development (I. Grant and C. Sear, eds.), 2 10-224. Natural Resources Institute, Chatham, UK.

Xie, I? and I? A. Arkin. 1997. A 17-year monthly analysis based

on gauge observations, satellite estimates, and numerical model outputs. Bull. Amer. Met. Soc. 78, 2539-2558.


Table 8.1. Some satellite sensors and data providers

Satellite Spatial Temporal Spectral Cost indication sensor resolution resolution bands per scene (approx.) Geostafionary satellites - for example for Africa: Meteosat 2.4-5km 30 min Visible, Free with receiving equipment

infrared, and license (variable cost, free water vapour to some users) from Eumetsal

MSG 1-3km 15 min 12 bands Polar orbiting satellites (low / medium resolution) - for example: NOAA-AVHRR 11 00m < 1 day 5 bands: red, Free with receiving equipment or

NIR, MIR, basically cost price if ordered 2xTIR

SPOT- 1150m 1 day 4 bands: blue, Contact SPOT image VEGETATION red, NIR & Free

SMIR Terra MODlS 250m 1-2 days 36 bands Free

500m [visible to 1000m infrared]

Polar orbiting sateflites (high resolution) - for example: SPOT 1 Om 26 days or less green, red, NIR Spotimage prices:

20m & SMlR 1250 - 5100 3m Panchromatic I

visible Landsat TM 30m 16 days 7 bands: blue, Landsat 5 TM: $ 2870

120m 1 60m green, red, Landsat 7 TM: $ 600 15m NIR, SMIR,

LMlR & TIR Panchromatic: 1 band: visible1 NIR

ERS SAR 12.5-30m 26 days Radar Eurimage: $ 1200 (discounts for multiple images)

ERS SAR 12.5-30111 26 days Radar Eurimage: $ 1200 (discounts for green, red, multiple images)

IRS 23m 24 days NIR, SWlR Spaceimaging Europe: 2500 5.8m Panchromatic/

visible l konos 4 m blue, green, Expensive (varies with product

I rn red, VNlR and quantity of data) Panchromatic: visible

Other useful sites Vegetation (NDVI) Free and Rainfall (CCD) Fire products Free

SAC

ADD

WFU lona See MOD

maPF

Dzta Access / $&lress


Contact / lnternet information

Euinetsat

Satellite Applications Centre (SAC) F.0 Box 395 Pretoria 0001

http://www.eumetsat.de/en/dps/helpdesk/ msg-suppliers. html

Tel: +27 (12) 334 5000 Fax: +27 (12) 334 5001 [email protected]

Republic of South Africa http://www.sac.co.za/ GSFC DAAC http://daac.gsfc.nasa.qov/CAMPAIGN DOCS1 Spot Image, France BRS-SRVR/avhrrbrs-main.html

http://free.vgt.vito.be/

EOSDIS http://redhook.gsfc.nasa.gov/-imswww/pub/ GES DAAC imswelcome/plain.htmI

http://daac.gsfc.nasa.gov/MODIS/ MODlS User Services: Phone: +1 (301) 614 5224 Fax: +1 (301) 61 4 5304 help @daac.gsfc.nasa.gov

SACSPOT image, France (see above)http://www.spot.com

SAC USGS (Landsat 7) (see above) [email protected] http://landsat7.usgs.gov/

Eurimage SAC (see above) http://earth.esa.int/helpandmail/ help-order. html

Eurimage SAC (see above) http://earth.esa.int/helpandmail/ help-order. html

Spaceimageing EOSAT csc@ si-eu.com http://www.spaceimaging.com/

SAC Space Imaging EOSAT (see above) http://www.spaceimaging.com/ defaukhtm

P.DDS African Dada Dissemination Service http://edcintl.cr.usgs.gov/adds

WFW World Fire Web http://www.gvm.jrc.it/TEM/wfw/wfw.htm lona Fire http://sharkl .esrin.esa.it/ionia/FIRE/ See also GFMC-remote sensing http://www.fire.uni-freiburg.de/inventory/ MODIS: Daily images and active fires Internet fire rem-pro. html mapping tool Daily active fire text files http://rapidfire.sci.gsfc.nasa.gov

http://maps.geog.umd.edu ftp://maps.geog.umd.edu

Date post:	14-Jul-2020
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

Remote sensing of vegetation fires and its contribution to ... · Remote sensing is the science and...

Documents